We’re thrilled to announce a major transformation of our cloud graph database offering, Neo4j AuraDB, making it dramatically easier for enterprises to deploy graph solutions in production for any workload or use case. The new features and capabilities include:

• Enterprise features at a lower cost: AuraDB Business Critical has high availability and advanced security for 20% less than AuraDB Enterprise, now renamed to AuraDB Virtual Dedicated Cloud
• Streamlined data management and user experience with Aura Console, which administers, models, and visualizes data across Neo4j tools, along with a new GenAI copilot to enhance productivity via a unified data management hub
• An interactive, low-code dashboard builder, NeoDash creates maps, bar charts, tables, and other visualizations
• Expanded vertical scaling with up to 512GB AuraDB instances for large, complex datasets and applications
• Ability to process up to 15x more data per cluster with read-only secondaries on AuraDB, enabling horizontal scaling without compromising latency or performance

We’re introducing these enhancements amid rising demand for Neo4j’s fully managed cloud offering. This demand is driven by graph databases’ critical role in developing GenAI apps that deliver highly accurate responses, rich context, deep explainability, advanced analytics applications, and the evolving data ecosystem.

According to Gartner, graph technologies will be used in 80% of data and analytics innovations by 2025, up from 10% in 2021.¹ Neo4j AuraDB will now allow enterprises to meet growing business needs for scalability, performance, sophisticated data management, and advanced analytics applications well into the future—even as data volumes and complexity continue to increase. The new features and capabilities, detailed below, are available now.

Introducing AuraDB Business Critical: Enterprise Features at a Lower Cost

AuraDB Business Critical is our new service tier designed for mission-critical applications that require advanced security, scalability, and round-the-clock support. At over 20% less than Neo4j AuraDB Virtual Dedicated Cloud, it provides a cost-effective option for organizations seeking enterprise-grade features. Business Critical enables both pay-as-you-go and prepaid consumption models, providing ultimate commercial flexibility for organizations.

With AuraDB Business Critical, you get:

• Highly available database cluster with 99.95% uptime SLA
• Built-in security with enhanced data protection and role-based access control
• Vertical scaling up to 512GB RAM AuraDB instances
• Daily backups with 30-day retention and hourly point-in-time restore
• Pay-as-you-go and prepaid consumption billing
• Regional 24 x 7 support

AuraDB Business Critical will soon be available in the same regions as AuraDB Professional over the next few quarters. For more information on AuraDB pricing and service level tiers, visit Neo4j pricing. Customers requiring dedicated infrastructure can continue to use AuraDB Enterprise, now renamed AuraDB Virtual Dedicated Cloud.

AuraDB Professional tier is now available as a free trial, allowing you to explore and test features before deploying workloads in production. The trial gives you access to a full-featured product to build and test with before making a purchase decision.

Streamlined Data Management and User Experience With Aura Console, Query API, GraphQL API, and Change Data Capture (CDC)

Aura Console, featuring a GenAI copilot, is a unified platform for developers and administrators that integrates Neo4j Bloom, Neo4j Browser, Neo4j Importer, and Neo4j OpsManager. This single interface allows users to efficiently administer, manage, ingest, model, and visualize data across Neo4j’s offerings and tools, enhancing usability and creating a streamlined user experience. With natural language interaction, the copilot assists in writing and improving Cypher queries and offers support for visually exploring graph data.

By simplifying and streamlining data management for developers and database administrators, Aura Console offers:

• Improved workflows with a single hub for all data management tasks
• Consistent user experience through a single UX across Neo4j tools
• Easier collaboration as teams can share resources and collaborate on projects
• Secure data access with expanded roles and new access controls
• Improve productivity and learning with a GenAI copilot by helping developers write and improve Cypher queries

To further improve accessibility and integration options, we’ve introduced two powerful API features:

Query API: Execute Cypher statements against your Neo4j server through simple HTTP requests. This new feature enables seamless communication between Neo4j and your other systems, making it easier to incorporate graph database capabilities into your existing workflows. Additionally, it enhances the integration of Neo4j with modern front-end technologies and frameworks, making it easier to build and iterate on applications, ultimately broadening the scope of what can be developed on the platform. Learn more about the Query API here.

GraphQL API: You can now enable GraphQL for AuraDB instances, connecting via this widely adopted and developer-friendly query language. With GraphQL, you can request exactly the data you need in a single API call, simplifying your data interactions. This enhancement also improves Neo4j’s integration with modern front-end technologies and frameworks, making it easier for you to build and iterate on applications. Ultimately, you’ll have a broader scope of what you can develop on the platform, unlocking new possibilities for your projects. Learn more about the GraphQL API here.

Aura Console: Various actions users can take along with querying the graph

To further streamline data management and user experience, we’ve also implemented the following capability:

Change Data Capture (CDC): Automates tracking and notification of database changes so organizations can make mission-critical decisions in real time. It also empowers application developers to create event-driven applications using Neo4j as the system of record, with the ability to filter for specific events. Our customers have built applications that include:

• Security systems  that  map access and entitlement
• Application governance and change management
• Building entity-resolution pipelines
• Active recommendation systems

We now offer Neo4j Connectors for Confluent and Apache Kafka v5.1, which include support for CDC strategies. Learn more about CDC and the latest connectors here.

Introducing NeoDash: Interactive, Low-Code Dashboards for Neo4j

NeoDash, a low-code dashboard builder, empowers users to easily create interactive visualizations of their graph data. Seamlessly integrated with Neo4j, NeoDash works natively with Cypher queries to build maps, charts, tables, and more, making data insights accessible to everyone.

NeoDash allows users to:

• Create visualizations with an intuitive drag-and-drop interface
• Build dashboards directly from Cypher queries
• Customize and add interactivity to their visualizations
• Publish dashboards for secure, read-only sharing with stakeholders

Neo4j now offers enterprise-level support for NeoDash, representing a significant milestone in its journey from a Neo4j Labs project to a fully integrated product. While the code remains in Labs, users can access official Neo4j Support and Professional Services to assist with their NeoDash deployments. As NeoDash continues to evolve, users can look forward to its full productization in 2025, further strengthening its position within the Neo4j ecosystem.

A sample Neo4j dashboard demonstrating how to build visual planners for rail networks and other complex systems.

Scale Up and Out With Larger AuraDB Instances and Read-Only Secondaries

We now offer AuraDB instances with up to 512GB of RAM, a 33% capacity increase, allowing you to work with more extensive graph datasets and accelerate complex queries and analytical tasks. These larger instances are ideal for memory-intensive workloads such as:

• High-frequency financial trading
• Processing and analyzing large-scale genomic data for biology research
• High-performance computing (HPC) workloads like complex climate models
• Modeling and optimizing complex supply chain networks
• Analyzing complex transaction patterns to identify fraudulent activities

In addition to vertical scaling, you can now horizontally scale your AuraDB instances with up to 15 Read-Only Secondaries. This new feature improves read performance and throughput by efficiently distributing read-heavy workloads across secondaries. Read-Only Secondaries are ideal for applications with high read-to-write ratios. They ensure consistent performance as your data grows by routing read queries to secondaries and non-leader primaries within the same region. Secondaries are a cost-effective way to scale read capacity without compromising on latency.

A Highly Secure, Comprehensive Graph Solution With Advanced Analytics to Meet Enterprise Needs

The new enhancements to Neo4j AuraDB complement the existing enterprise security features, creating a comprehensive solution for GenAI and advanced analytics. We offer advanced security features, including Customer Managed Keys (CMK) and Security Log Forwarding.

CMK gives customers full control over their encryption keys, including policies, rotation, and versions, ensuring that only authorized users have access. CMK helps organizations meet essential regulatory and industry standards, including GDPR and HIPAA, and it’s now supported on any cloud provider.

Creating Customer Managed Key with Encryption Key ARN.

Security Log Forwarding allows users to stream security logs in real time to a cloud project owned by their organization. This enhances threat detection and response by providing a unified view of the organization’s security posture.

Users can set up and configure Security Log Forwarding for AuraDB, where the destination is a cloud logging service. The supported targets are Google Cloud Logging in GCP, Cloudwatch in AWS, and Log Analytics in Microsoft Azure. A separate log-forwarding process can be defined per cloud provider and region if multiple regions are in use.

The latest AuraDB enhancements reflect Neo4j’s ongoing commitment to our product vision: to deliver an easy-to-use, premium, and trusted cloud-native graph database and analytics offering that’s available on any cloud, ultimately enabling developers, data analysts, and data scientists to build accurate and explainable AI applications.

These innovations build on recent efforts, which include:

• Partnering with Snowflake to bring graph data science into the AI data cloud
• Collaborating with Microsoft to natively integrate Neo4j’s graph capabilities into the Microsoft Fabric analytics platform and Microsoft Azure OpenAI Service.
• Bringing graphRAG capabilities for GenAI to Google Cloud
• Introducing new resources for GenAI applications, like the Neo4j LLM Knowledge Graph Builder.

With this momentum, we’re ready to accelerate the pace of graph technology breakthroughs even further in the year ahead.

Start your graph journey with AuraDB today and experience the benefits of Neo4j’s enterprise-ready graph technology.

Check out our comprehensive guide, explore hands-on training on GraphAcademy, engage with our vibrant developer community, and access drivers and APIs to accelerate your graph success.

Other Resources

• AuraDB Page
• AuraDB Documentation
• Developer Center
• AuraDB Pricing Plans
• Webinar: Accelerate Your Workloads with AuraDB’s Latest Enhancements

¹ Gartner, Emerging Tech Impact Radar: Data and Analytics, By Kevin Quinn, Radu Miclaus, Ben Fieselmann, et al, November 20, 2023

The rise of generative AI (GenAI) is transforming how businesses operate and society’s very fabric. With the potential to alter how we work, create, communicate, and consume, GenAI is poised to have an impact as significant as the internet and the mobile phone (combined!).

GenAI will bring many opportunities and challenges as it advances and permeates various aspects of our lives. GenAI has the power to automate mundane tasks and unlock new possibilities for creativity and problem-solving — from personalized content creation to data-driven decision-making.

However, the rapid growth of GenAI also raises important questions about its potential unintended consequences. As the technology becomes more sophisticated and ubiquitous, it’s critical to consider data privacy, algorithmic bias, and the ethical implications of AI-generated content. Developers, tech leaders, and policymakers will face a crucial challenge in ensuring responsible development and deployment of GenAI in the coming years.

For those working in data and AI, staying informed about the capabilities and implications of GenAI is essential. Whether you’re a developer building AI-powered applications, a data scientist exploring new use cases for GenAI, or a tech leader guiding your organization’s AI strategy, a deep understanding of this transformative technology will be critical to success in the era of GenAI.

In this blog post, we provide a comprehensive introduction to GenAI, covering its key characteristics, the rise of large language models (LLMs), and how it differs from traditional AI.

How Does Generative AI Work?

Generative AI, or GenAI, is short for generative artificial intelligence. GenAI is a subset of artificial intelligence (AI) that creates new content based on learned patterns and rules, such as text, images, audio, and code. Unlike traditional AI systems designed to recognize or classify existing data, generative AI models can produce novel outputs that resemble the data they were trained on.

Generative AI works by using advanced machine learning algorithms, particularly deep learning and neural networks, to analyze and learn from large datasets. Transformer models are at the core of many generative AI systems and use self-attention mechanisms to process and generate sequential data.

The Rise of the Transformer Model

The foundation of modern GenAI can be traced back to the introduction of the transformer model in a 2017 paper titled “Attention Is All You Need” by researchers at Google Brain. The transformer model revolutionized the field of natural language processing (NLP) by introducing a self-attention mechanism that allowed the model to weigh the importance of different parts of the input when generating output.

Transformers convert text and data into vector embeddings, which are numerical representations of the input. With the help of NLP, the vectorization process helps determine the semantics, such as how words are related, and dictates the ability of the generative AI model to produce output similar (though not identical) to its training data.

The transformer architecture and other deep learning techniques enabled the development of large language models trained on vast amounts of text data to generate human-like responses. One of the first notable examples of a large language model was GPT (Generative Pre-trained Transformer), developed by OpenAI in 2018. GPT showcased the potential of transformer-based models for generating coherent and contextually relevant text.

The Era of LLMs

Following the success of GPT, several other LLMs were introduced, each pushing the boundaries of what was possible with GenAI. Models like Google’s BERT (Bidirectional Encoder Representations from Transformers) and Facebook AI Research’s RoBERTa (Robustly Optimized BERT Pretraining Approach) demonstrated remarkable performance on various NLP tasks, from sentiment analysis to question answering.

As LLMs grew in size and complexity, they exhibited increasingly impressive capabilities. Models like GPT-3, released by OpenAI in 2020, showcased the ability to generate highly coherent and contextually relevant text with minimal prompting. This marked a significant milestone in the evolution of GenAI, as it demonstrated the potential for models to generate human-like content with unprecedented fluency. Researchers also explored techniques like fine-tuning to adapt pre-trained models to specific domains or tasks, further enhancing their performance and versatility. Now, large language models like GPT-4 and DALL-E 3 can generate human-like text, realistic images, and even code.

The ChatGPT Phenomenon

In November 2022, OpenAI released ChatGPT, a conversational AI model based on GPT-3.5. ChatGPT quickly gained widespread attention for its ability to engage in human-like conversations, answer questions, and generate text across various topics. The model’s ease of use and impressive output quality made it accessible to a broad audience, sparking a surge of interest in GenAI among the general public.

The success of ChatGPT brought GenAI into the mainstream, highlighting the potential for these models to transform various industries and aspects of our daily lives. It also raised important questions about the ethical implications of AI-generated content, the potential for misuse, and the need for responsible development and deployment of GenAI technologies.

How Generative AI Differs from Traditional Artificial Intelligence

Traditional AI typically involves algorithms designed to perform specific tasks. These models are trained on predefined datasets and operate within a fixed scope, following set rules or patterns. For instance, a traditional AI model might excel at classifying images or predicting stock prices, but its functionality is limited to those particular tasks. It requires structured data and extensive manual feature engineering to achieve high accuracy.

Generative AI models, like GPT, are trained on vast datasets and can generate new content across various domains, from text and images to code and beyond. The key distinction lies in GenAI’s ability to synthesize novel ideas and solve complex problems that mimic human creativity. This flexibility makes GenAI a powerful asset for tasks requiring innovation and adaptability, going beyond the limitations of traditional AI’s specialized functionality.

What Are the Benefits of Generative AI?

GenAI presents opportunities to improve decision-making, customer experiences, operational efficiency, and innovation:

1. Strategic Decision-Making: GenAI can analyze vast amounts of structured and unstructured data, providing insights that inform strategic decision-making. By generating comprehensive reports, identifying trends, and predicting potential outcomes, GenAI empowers business leaders to make data-driven decisions with confidence.
2. Enhanced Personalization: GenAI enables businesses to personalize customer interactions at scale. From generating tailored content recommendations to providing intelligent customer support through AI chatbots and conversational AI, GenAI helps businesses deliver exceptional experiences that foster customer loyalty and satisfaction.
3. Operational Efficiency: GenAI can significantly improve operational efficiency by automating routine tasks and streamlining processes, like optimizing supply chain management or generating financial reports. This frees human resources to focus on higher-value activities, resulting in cost savings and increased productivity.
4. Innovation and Competitive Advantage: GenAI enables businesses to explore new opportunities and stay ahead of the competition. By assisting humans in generating ideas, designs, and solutions, GenAI can accelerate product development, identify untapped markets, and create differentiated offerings that set businesses apart in their industries.

The examples mentioned above merely scratch the surface of what’s possible with GenAI. As the technology continues to evolve, we can expect to see even more innovative applications across industries, from healthcare and finance to manufacturing and beyond.

However, to apply GenAI to mission-critical systems, businesses must navigate various challenges and pitfalls. While GenAI offers immense opportunities, it also presents unique risks and considerations that must be addressed to ensure its responsible and effective deployment.

Popular Generative AI Use Cases

As GenAI evolves, developers are building a wide array of generative AI tools that can be applied across various industries. These tools are expanding the potential applications of GenAI, including:

Enhanced Customer Experiences

As mentioned earlier, GenAI can be used to personalize customer experiences. More specific customer interactions and marketing strategies include:

• Real-time Personalization: GenAI models can instantly process user data such as demographics, behavior, and context to identify patterns and preferences, generating highly personalized marketing copy and visuals on the fly.
• Chatbots and Virtual Agents: Next-generation AI-powered chatbots can provide more natural, context-aware responses and even initiate actions on behalf of customers.

Software Development and Building Applications

GenAI can accelerate software development processes and building apps through:

• Automated Code Generation: AI tools can write and optimize code, speeding up development and reducing errors.
• Application Modernization: GenAI can automate the repetitive coding required to update legacy applications for modern cloud environments.
• Bug Detection and Fixing: AI algorithms can analyze codebases to identify potential bugs and vulnerabilities, as well as suggest optimizations for improved performance.

Productivity and Workflow Optimization

GenAI can enhance operational efficiency across various business functions:

• Automated Documentation: GenAI can generate and summarize business proposals, legal documents, reports, and other corporate communications.
• Digital Labor: AI can assist in creating or revising contracts, invoices, and other paperwork, freeing up employees for higher-value tasks.
• Intelligent Assistance: GenAI can provide context-aware support for employees handling corporate and customer information, boosting productivity.

Product Design and Innovation

GenAI is shifting the product development process:

• Design Iteration: AI algorithms can generate and evaluate multiple design options, helping engineers and designers create better products more efficiently.
• Simulation and Analysis: GenAI can run complex simulations, conduct what-if scenarios, and assess risks, providing valuable insights for product development and optimization.

Advanced Analytics and Decision Support

GenAI is enhancing data analysis and decision-making capabilities:

• Predictive Analytics: AI models can analyze vast amounts of data to identify trends, forecast outcomes, and provide actionable insights.
• Risk Assessment: GenAI can evaluate complex scenarios and compile findings into comprehensive reports, supporting informed decision-making.

Scientific Research and Drug Discovery

GenAI is accelerating scientific breakthroughs:

• Data Analysis: AI algorithms can process and analyze large volumes of data from clinical trials, scientific literature, and other sources to discover and identify potential drug targets or research directions.
• Image Generation: In healthcare, generative models can create synthetic medical images for training and testing imaging systems.

Content Generation and Creative Assistance

GenAI can augment creative processes:

• Automated Content Creation: AI can generate various types of content, including articles, social media posts, and even basic video scripts. More and more people are also using AI for image generation.
• Creative Ideation: GenAI can assist in brainstorming sessions, providing novel ideas and perspectives to spark human creativity.

Generative AI Challenges

Generative AI models are only as good as the data on which they are trained. Biased, inaccurate, or incomplete data can lead to flawed outputs and perpetuate unfairness.

Data Quality and Bias

The quality and diversity of training data are critical factors in the performance and fairness of GenAI models. Biased, inaccurate, or incomplete data can lead to flawed outputs, perpetuate unfairness, and even amplify societal biases. For example, if a GenAI model is trained on a dataset that underrepresents certain demographics, it may generate outputs that are less accurate or relevant for those groups.

AI companies are actively working to address these challenges. OpenAI, Google, and Anthropic have all emphasized the importance of curating diverse and representative datasets. For example, Anthropic developed a technique called “constitutional AI” to align their models with specific values, reduce the risk of harmful outputs, and ensure that the model’s outputs align with ethical principles. Techniques like generative adversarial networks (GANs) and diffusion models are also being explored to generate more diverse and representative training data, helping to mitigate biases in GenAI models.

Accuracy and Hallucinations

Even with high-quality training data, GenAI models can sometimes generate inaccurate or nonsensical outputs, known as “hallucinations.” These hallucinations can range from minor inconsistencies to completely fabricated information. In mission-critical applications, such inaccuracies can have serious consequences, leading to incorrect decisions or actions.

Addressing the issue of hallucinations is an ongoing challenge in the GenAI field. Researchers are exploring techniques like “consistency training” and “factuality scoring” to improve the accuracy and coherence of generated outputs. However, achieving perfect accuracy remains an elusive goal, and businesses must be aware of the limitations and potential risks associated with GenAI systems.

Explainability and Transparency

The inner workings of GenAI models are often described as a “black box.” Due to the complexity and scale of GenAI models, it can be challenging to understand exactly how they arrive at their outputs. This lack of transparency raises concerns about accountability and trust, particularly in mission-critical applications where the stakes are high.

Even industry leaders like Google have acknowledged the challenges of explainability in AI. In a 60 Minutes interview, Google CEO Sundar Pichai admitted that people “don’t fully understand” how artificial intelligence works and the state of the technology is still somewhat of a black box to researchers.

Efforts are underway to improve the explainability and transparency of GenAI models. Techniques like “attention visualization” and “probing” aim to provide insights into the model’s decision-making process. Additionally, some companies, like Anthropic, are taking novel approaches to peek inside the black box.

Anthropic researchers are trying to reverse engineer their language model, Claude, to understand why it generates specific outputs. By identifying combinations of artificial neurons that evoke specific concepts or “features,” they have created a sort of Rosetta Stone to decode the neural network. This approach has the potential to make LLMs safer by pinpointing where dangers lurk within the model.

Security and Privacy

The use of GenAI in mission-critical systems also raises concerns about security and privacy. GenAI models often require access to sensitive data, such as personal information or proprietary business data. Ensuring the security and confidentiality of this data is crucial to maintaining trust and complying with regulations.

Businesses must implement robust security measures to protect their GenAI systems from unauthorized access, data breaches, and malicious attacks. This includes techniques like encryption, access controls, and secure computing environments. Additionally, privacy-preserving techniques like differential privacy and federated learning can help to protect individual data points while still enabling the benefits of GenAI.

Emerging Techniques to Address Generative AI Challenges

As businesses explore the potential of GenAI, they are grappling with the challenges of accuracy, explainability, and security, especially when pairing GenAI with their proprietary data. To address these challenges, particularly in enterprise settings, new techniques are emerging that combine large language models with other data systems and data sources, particularly in enterprise settings.

One such technique is Retrieval-Augmented Generation (RAG), which enhances the responses of language models by retrieving relevant information from external data sources. RAG enables GenAI systems to provide more accurate and contextually relevant answers by grounding the generated content in verified information. It’s like giving the AI access to a reference library to improve its answers.

Taking this a step further, GraphRAG (Graph-Based Retrieval-Augmented Generation) uses knowledge graphs to provide an even richer context for GenAI responses. By representing data in a structured and interconnected manner, knowledge graphs enable GenAI systems to understand complex relationships, perform multi-hop reasoning, and generate more accurate and explainable outputs.

Example of a knowledge graph.

GraphRAG offers several key benefits for enterprise GenAI applications:

1. Improved Accuracy: By integrating domain-specific knowledge from enterprise data sources, GraphRAG reduces the risk of hallucinations and ensures that generated content aligns with verified facts.
2. Enhanced Relevancy and Context: Knowledge graphs enable GenAI systems to understand the relationships between entities and concepts, providing a rich context for generating relevant and meaningful responses. By tapping into the interconnected nature of knowledge graphs, GraphRAG can surface insights and recommendations that are tailored to the user’s specific needs and context.
3. Increased Explainability: Knowledge graphs provide a transparent structure that allows users to trace the reasoning behind GenAI outputs, making the decision-making process more understandable and auditable. By visualizing the connections between entities and concepts, GraphRAG enables users to understand how the AI arrived at a particular conclusion, increasing trust and accountability.

As these techniques mature, we can expect to see more enterprises adopting GraphRAG and similar approaches to build trustworthy, context-aware GenAI applications that drive real business value.

The Future of GenAI: Unlocking New Possibilities

The rapid advancements in GenAI are just the beginning of a transformative journey. As the technology continues to evolve, we can anticipate even more impressive capabilities and applications across industries.

In the near future, GenAI will likely become an integral part of our daily lives, seamlessly integrating with various devices and platforms to provide personalized, context-aware assistance. From smart homes to virtual personal assistants, GenAI will play a crucial role in shaping our interactions with technology.

For businesses, GenAI will open up new avenues for innovation, efficiency, and growth. Organizations will be able to automate complex tasks, gain deeper insights from vast amounts of data, and create highly personalized customer experiences. As the data ecosystem around GenAI matures, even small and medium-sized enterprises will be able to tap into its potential to level the playing field and compete with larger players. Expectations are rosy on that note, and experts predict that GenAI will be pervasive within a decade:

• By 2025, it’s expected to be in 80% of conversational AI tools
• By 2030, in part due to GenAI (McKinsey), activity that accounts for about a third of US workers’ time could be automated
• By 2032, the GenAI market is expected to hit $191 billion

However, the future of GenAI also demands a responsible and ethical approach. As GenAI becomes more powerful and pervasive, it is crucial to prioritize transparency, accountability, and fairness. Developers, policymakers, and business leaders must work together to establish guidelines and best practices that ensure the safe and beneficial deployment of GenAI systems.

At Neo4j, we are committed to empowering businesses to build accurate and explainable GenAI applications. Our GraphRAG approach combines the power of knowledge graphs, vector search, and generative AI models to create accurate, explainable, and context-rich GenAI solutions.

A knowledge graph is an organized representation of real-world entities and their relationships. It is typically stored in a graph database, which natively stores the relationships between data entities. Entities in a knowledge graph can represent objects, events, situations, or concepts. The relationships between these entities capture the context and meaning of how they are connected.

A knowledge graph stores data and relationships alongside frameworks known as organizing principles. They can be thought of as rules or categories around the data that provide a flexible, conceptual structure to drive deeper data insights. The usefulness of a knowledge graph lies in the way it organizes the principles, data, and relationships to surface new knowledge for your user or business. The design is useful for many usage patterns, including real-time applications, search and discovery, and grounding generative AI for question-answering.

Sometimes, people overcomplicate the concept of a knowledge graph. You might hear about enterprise-wide structures that consolidate and connect information across data silos and various sources. While that does describe a knowledge graph (one that can underpin a data integration use case), it describes one with a wide scope. Thinking only in terms of bridging large datasets and multiple data sources can make creating and implementing knowledge graphs seem complicated and time-consuming. But knowledge graphs don’t need to be broad or elaborate. You can build one with a much smaller scope to solve a use-case-specific problem.

How Knowledge Graphs Work

You may have heard of knowledge graphs in the context of search engines. The Google Knowledge Graph changed how we search for and find information on the Web. It amasses facts about people, places, and things into an organized network of entities. When you do a Google search for information, it uses the connections between entities to surface the most relevant results in context, for example, in the box Google calls the “knowledge panel.”

The Google knowledge panel of La Sagrada Familia includes an image of the site, a map, a description, address, hours of operation, the architects who built it, its height, and more.

The entities in the Google knowledge graph represent the world as we know it, marking a shift from “strings to things.” Behind this simple phrase is the profound concept of treating information on the web as entities rather than a bunch of text. Since information is organized as a network of entities, Google can tap into the collective intelligence of the knowledge graph to return results tailored to the meaning of your query rather than a simple keyword match.

Key Characteristics

Now that you understand how knowledge graphs organize and access data with context, let’s look at the building blocks of a knowledge graph data model. The definition of knowledge graphs varies depending on whom you ask, but we can distill the essence into three key components: nodes, relationships, and organizing principles.

Nodes

Nodes denote and store details about entities, such as people, places, objects, or institutions. Each node has a (or sometimes several) label to identify the node type and may optionally have one or more properties (attributes). Nodes are also sometimes called vertices.

For example, the nodes in an e-commerce knowledge graph typically represent entities such as people (customers and prospects), products, and orders:

Relationships

Relationships link two nodes together: they show how the entities are related. Like nodes, each relationship has a label identifying the relationship type and may optionally have one or more properties. Relationships are also sometimes called edges.

In the e-commerce example, relationships exist between the customer and order nodes, capturing the “placed order” relationship between customers and their orders:

Organizing Principle(s)

Organizing Principles are a framework, or schema, that organizes nodes and relationships according to fundamental concepts essential to the use cases at hand. Unlike many data designs, knowledge graphs easily incorporate multiple organizing principles.

Organizing principles range from simple (product line -> product category -> product taxonomy) to complex (a complete business vocabulary that explains the data in the graph). Think of an organizing principle as a conceptual map or metadata layer overlaying the data and relationships in the graph.

The model uses the same node-and-relationship structure as the rest of the knowledge graph to describe the organizing principles – which means you can write queries that draw from both instance data and organizing principles.

In the e-commerce example, an organizing principle might be product types and categories:

What About Ontologies?

When learning about knowledge graphs, you might come across articles on ontologies and wonder where they fit in. An ontology is a formal specification of the concepts and the relationships between them for a given subject area; semantic networks are a common way to represent ontologies. Put simply, ontologies are a type of organizing principle.

Ontologies can be complex and require a great deal of effort to define and maintain. When deciding whether an ontology is needed, it’s critical to consider the problems you’re trying to solve with a knowledge graph. In many cases, it won’t be necessary. In the e-commerce example, using a product taxonomy as the organizing principle is sufficient for a product recommendation use case.

Think of the knowledge graph as a growing, evolving system to simplify your design in the early stages and deliver value sooner. If you pick the right technology to implement your knowledge graph, you can expand and evolve the graph as your needs change. In this way, you can add ontologies when your use case requires them rather than forcing yourself to build them up-front.

Knowledge Graph Example

Let’s see what a knowledge graph might look like. Below is a simple knowledge graph of the e-commerce example that shows nodes as circles and relationships between them as arrows. The organizing principles are also stored as nodes and relationships, so the illustration uses some color shading to show which nodes and relationships are the instance data and which are the organizing principles:

An example knowledge graph showing nodes as circles and relationships as arrows. The instance data and organizing principles are highlighted for display.

Knowledge Graphs and Graph Databases

Creating a knowledge graph involves conceptually mapping the graph data model and then implementing it in a database. There are many databases to choose from, but choosing the right one can simplify the design process, speed up development and implementation, and make it easier to adapt to future changes and improvements.

Property Graphs

Native property graph databases, such as Neo4j, are a logical choice for implementing knowledge graphs. They natively store information as nodes, relationships, and properties, allowing for an intuitive visualization of highly interconnected data structures. The physical database matches the conceptual data model, making designing and developing the knowledge graph easier. When you use property graphs, you get:

• Simplicity and ease of design: Property graphs allow for straightforward data modeling when designing the knowledge graph. Because the conceptual and physical models are very similar (often the same), the transition from design to implementation is more straightforward (and easy to explain to non-technical users).


• Flexibility: It’s easy to add new data, properties, relationship types, and organizing principles without extensive refactoring or code rewrites. As needs change, you can iterate and incrementally expand the knowledge graph’s data, relationships, and organization.


• Performance: Property graphs offer superior query performance compared to alternatives like RDF databases or relational databases, especially for complex traversals and many-to-many relationships. This performance comes from storing the relationships between entities directly in the database rather than re-generating them using joins in queries. A native property graph database traverses relationships by following pointers in memory, making queries that traverse even complex chains of many relationships very fast.


• Developer-friendly Code: Property graphs support an intuitive and expressive ISO query language standard, GQL, which means you have less code to write, debug, and maintain than SQL or SPARQL. Neo4j’s Cypher is the most widely used implementation of GQL.

Property Graph Vs. Triple Stores (RDF)

People sometimes think of property graphs and triple stores as equally viable options for building a knowledge graph, but triple stores (also known as RDF databases) have considerable disadvantages.

Based on the Resource Description Framework (RDF), triple stores use a granular approach to design and storage. Triple stores express all data in the form of subject-predicate-object “triples.” This model does not support relationships with properties or multiple same-typed relationships between entities. To accommodate real-world use cases, you will need to implement workarounds. Common workarounds include turning relationships into objects (called reification) or using singleton properties to capture properties using extra “type-of” relationships. These workarounds mean larger databases, additional complexity in the physical model, and poor query performance.

Because reification and singleton properties force tough decisions about the design up front, triple stores don’t lend themselves to solving real-world problems that involve messy data domains. Knowledge graphs built on a triple store are more challenging to design, time-consuming to implement, and difficult to change.

Property Graph Vs. Relational Databases

Relational databases and other non-native graph approaches suffer similar design friction. Neither relational nor document databases store relationships – they must be synthesized at runtime with joins or value lookups in query code. Since the relationships reside in the code rather than with the dataset, each application and data use must have its own implementation. SQL (the relational database query language) forces you to define every join in the query itself. As a result, the knowledge graph becomes more difficult to manage and yields poor runtime performance as the number of relationships expands.

Knowledge Graph Use Cases

Knowledge graphs offer a powerful tool for storing and organizing data to enable a more sophisticated understanding of that data. To understand how companies have done this, let’s look at examples of using knowledge graphs to tackle particular problems. Though not a comprehensive list of use cases, it’s a set of concrete examples demonstrating knowledge graphs in real-world applications.

Generative AI for Enterprise Search Applications

In generative AI applications, knowledge graphs capture and organize key domain-specific or proprietary company information. Knowledge graphs are not limited to structured data; they can handle less organized data as well.

GraphRAG, a technique that grounds large language models with knowledge graphs, is emerging as the foundation of AI applications that use proprietary domain data (these are known as RAG applications). A knowledge graph grounding increases response accuracy and improves explainability with the context provided by data relationships. Industry leaders such as Deloitte highlight the critical role of knowledge graphs for building enterprise-grade GenAI. Gartner places knowledge graphs having a “high mass,” being an impactful technology for GenAI today:

This Impact Radar from Gartner highlights knowledge graphs as a high-impact technology within the Generative AI landscape (Credit: Gartner)

Fraud Detection and Analytics in Financial Services, Banking, and Insurance

In Fraud Detection and Analytics, the knowledge graph represents a network of transactions, their participants, and relevant information about them. Companies can use this knowledge graph to quickly identify suspicious activity, investigate suspected fraud, and evolve their knowledge graph to keep up with changing fraud patterns. Algorithms such as pathfinding and community detection provide key signals to machine learning algorithms that can uncover more sophisticated fraud networks.

Master Data Management

In Master Data Management (e.g., for Customer 360 use cases), the knowledge graph provides an organized, resolved (i.e., “de-duped”), and comprehensive database of a company’s customers and the company’s interactions with them.

This organized view of customers is especially important for companies with multiple divisions or applications interacting with customers. Without a knowledge graph, it can be difficult or impossible to obtain an accurate view of the customer. A knowledge graph links customer behaviors across multiple applications through an organizing principle that identifies them as coming from the same customer.

Supply Chain Management

In Supply Chain Management, a knowledge graph represents the network of suppliers, raw materials, products, and logistics that work together to supply a company’s operations and customers. This end-to-end supply chain visibility allows managers to identify weak points and predict where disruptions may occur. Graph algorithms such as shortest path optimize the supply chain in real time by finding the most direct route between A and B.

Investigative Journalism

In Investigative Journalism, knowledge graphs capture key entities (companies, people, bank accounts, etc.) and activities under investigation. Organizing these entities in relation to one another makes it possible to find hidden patterns, such as distant relationships between entities that shouldn’t be present.

Investigators may use techniques such as entity resolution to reveal entities hiding behind fake or shell identities to mask their activities. Algorithms like community detection and link prediction also provide insight and areas for further investigation.

Drug Discovery in Healthcare Research

Knowledge graphs store information about the research subject in medical and other research use cases. For example, the knowledge graph could have protein and genome sequences together with environmental and chemical data, revealing intricate patterns and expanding our knowledge of proteins.

Getting Started With Knowledge Graphs

Knowledge graphs are organized representations of real-world entities and their relationships, overlaid with one or more organizing principles that frame the information in context to drive insight from the data. Knowledge graphs underpin insightful applications and artificial intelligence solutions across many use cases.

To master the concepts and techniques behind knowledge graphs and get hands-on experience, download a free copy of the O’Reilly book Building Knowledge Graphs: A Practitioner’s Guide by Jesús Barrasa and Jim Webber.

The guide covers how to build, manage, query, analyze, and visualize your knowledge graph so you can develop data-backed applications and advanced analytics.

Get My Copy

Learning Resources

• Semantics workshops on GitHub.
• Ontology-Based Reasoning 101.
• Ontology-driven Knowledge Graph Construction.
• GraphAcademy for knowledge graph fundamentals using a property graph model in the Neo4j Graph Database.

We’re Entering the “Blue Links” Era of RAG

We are on the verge of realizing that in order to do anything significantly useful with GenAI, you can’t depend only on autoregressive LLMs to make your decisions. I know what you’re thinking: “RAG is the answer.” Or fine-tuning, or GPT-5.

Yes. Techniques like vector-based RAG and fine-tuning can help. And they are good enough for some use cases. But there’s another whole class of use cases where these techniques all bump into a ceiling. Vector-based RAG – in the same way as fine-tuning – increases the probability of a correct answer for many kinds of questions. However neither technique provides the certainty of a correct answer. Oftentimes they also lack context, color, and a connection to what you know to be true. Further, these tools don’t leave you with many clues about why they made a particular decision.

Back in 2012, Google introduced their second-generation search engine with an iconic blog post titled “Introducing the Knowledge Graph: things, not strings¹.” They discovered that a huge leap in capability is possible if you use a knowledge graph to organize the things represented by the strings in all these web pages, in addition to also doing all of the string processing. We are seeing this same pattern unfold in GenAI today. Many GenAI projects are bumping up against a ceiling, where the quality of results is gated by the fact that the solutions in use are dealing in strings, not things.

Fast forward to today, AI engineers and academic researchers at the leading edge are discovering the same thing that Google did: that the secret to breaking through this ceiling is knowledge graphs. In other words, bring knowledge about things into the mix of statistically-based text techniques. This works just like any other type of RAG, except with a call to a knowledge graph in addition to a vector index. Or in other words, GraphRAG!

This post is intended to be a comprehensive and easy-to-read treatment of GraphRAG. It turns out that building a knowledge graph of your data and using it in RAG gives you several powerful advantages. There’s a robust body of research proving that it gives you better answers to most if not ALL questions you might ask an LLM using normal vector-only RAG.

That alone will be a huge driver of GraphRAG adoption. In addition to that, you get easier development thanks to data being visible when building your app. A third major advantage is that graphs can be readily understood and reasoned upon by humans as well as machines. Building with GraphRAG is therefore easier, gives you better results, and – this is a killer in many industries – is explainable and auditable! I believe GraphRAG will subsume vector-only RAG and emerge as the default RAG architecture for most use cases. This post explains why.

Wait, Graph?

Let’s be clear that when we say graph, we mean something like this:

While this image has been widely used to exemplify knowledge graphs, the original source and author remain unidentified. The earliest known usage appears to be this Medium post from Farahnaz Akrami. If you are the creator of this image, please contact us so we may provide proper attribution.

The Graph of Thrones visualization by William Lyon.

London Underground Map (Credit: Transport for London.) Fun fact, Transport for London recently deployed a graph-powered digital twin to improve incident response and reduce congestion.

In other words, not a chart.

If you want to delve more into graphs and knowledge graphs, I’d recommend a detour to Neo4j’s GraphAcademy or Andrew Ng’s Deeplearning.ai course on Knowledge Graphs for RAG. We won’t linger on definitions here and will continue forward assuming basic working knowledge of graphs.

If you understand the pictures above, you can see how you might query the underlying knowledge graph data (stored in a graph database) as part of your RAG pipeline. This is what GraphRAG is about.

Two Types of Knowledge Representation: Vectors & Graphs

The core of typical RAG – vector search – takes in a chunk of text and returns conceptually similar text from a candidate body of written material. This is pleasantly automagical and is very useful for basic searches.

What you might not think about every time you do this is what a vector looks like, or what the similarity calculation is doing. Let’s look at an apple in human terms, vector terms, and graph terms:

The human representation is complex and multidimensional and not something we can fully capture on paper. Let’s grant some poetic license and imagine that this beautifully tempting picture represents an apple in all its perceptual & conceptual glory.

The vector representation of the apple² is an array of numbers – a construct of the statistical realm. The magic of vectors is that they each capture the essence of their corresponding text in encoded form. In a RAG context however, they are only valuable when you need to identify how similar one handful of words is to another. Doing this is as simple as running a similarity calculation (aka vector math) and getting a match. However, if you want to make sense of what’s inside of a vector, understand what’s around it, get a handle on the things represented in your text, or understand how any of these fit into a larger context, then vectors as a representation just aren’t able to do that.

Knowledge graphs, by contrast, are declarative – or in AI terms, symbolic – representations of the world. As a result, both humans and machines can understand and reason upon knowledge graphs. This is a BIG DEAL, which we’ll revisit later. Additionally, you can query, visualize, annotate, fix, and grow knowledge graphs. A knowledge graph represents your world model³ – the part of the world that represents the domain you are working with.

GraphRAG “vs.” RAG

It’s not a competition Vector and graph queries each add value in RAG. As pointed out by founder of LlamaIndex Jerry Liu, it’s helpful to think about GraphRAG as inclusive of vectors. This is distinct from “vector-only RAG,” which is strictly based on similarity with embeddings based on words in text.

Fundamentally, GraphRAG is RAG, where the Retrieval path includes a knowledge graph. As you can see below, the core GraphRAG pattern is straightforward. It’s basically the same architecture as RAG with vectors⁴ but with a knowledge graph layered into the picture.

GraphRAG Pattern

Here, you see a graph query being triggered. It can optionally include a vector similarity component. You can choose to store your graphs and vectors either separately in two distinct databases, or use a graph database like Neo4j which also supports vector search.

One of the common patterns for using GraphRAG is as follows:

1. Do a vector or keyword search to find an initial set of nodes.
2. Traverse the graph to bring back information about related nodes.
3. Optionally, re-rank documents using a graph-based ranking algorithm such as PageRank.

Patterns vary by use case, and like everything else in AI today, GraphRAG is proving to be a rich space, with new discoveries emerging every week. We will dedicate a future blog post to the most common GraphRAG patterns we see today.

GraphRAG Lifecycle

A GenAI application that uses GraphRAG follows the same pattern as any RAG application, with an added “create graph” step at the start:

Creating a graph is analogous to chunking documents and loading them into a vector database. Advances in tooling have made graph creation literally that easy. The good news is threefold:

1. Graphs are highly iterative – you can start with a “minimum viable graph” and expand from there.
2. Once your data is in a knowledge graph, it becomes very easy to evolve. You can add more kinds of data, to reap the benefits of data network effects. You can also improve the quality of the data to up the value of your application results.
3. This part of the stack is rapidly improving, which means graph creation will only get easier as tooling gets more sophisticated.

Adding the graph creation step to the earlier picture gives you a pipeline that looks like this:

I will dive deeper into graph creation later. For now, let’s set that aside and talk about the benefits of GraphRAG.

Why GraphRAG?

The benefits we are seeing from GraphRAG relative to vector-only RAG fall into three main buckets:

1. Higher accuracy and more complete answers (runtime / production benefit)
2. Once you’ve created your knowledge graph, then it’s easier to both build⁵ and subsequently maintain your RAG application (development time benefit)
3. Better explainability, traceability⁶, and access controls (governance benefit)

Let’s drill into these:

#1: Higher Accuracy & More Useful Answers

The first (and most immediately tangible) benefit we see with GraphRAG is higher-quality responses. In addition to a growing number of examples we see from our customers, an increasing number of academic studies also support this. One such example is by data catalog company Data.world. At the end of 2023, they published a study that showed that GraphRAG, on average, improved accuracy of LLM responses by 3x across 43 business questions. The benchmark found evidence of a significant improvement in the accuracy of responses when backed by a knowledge graph.

More recently and perhaps better known is a series of posts by Microsoft starting in February 2024 with a research blog titled GraphRAG: Unlocking LLM discovery on narrative private data, along with an associated research paper, and software release. Here they observed that baseline RAG (i.e. with vectors) has the two following problems:

• Baseline RAG struggles to connect the dots. This happens when answering a question requires traversing disparate pieces of information through their shared attributes in order to provide new synthesized insights.
• Baseline RAG performs poorly when being asked to holistically understand summarized semantic concepts over large data collections or even singular large documents.

Microsoft found that “By using the LLM-generated knowledge graph, GraphRAG vastly improves the ‘retrieval’ portion of RAG, populating the context window with higher relevance content, resulting in better answers and capturing evidence provenance.” They also discovered that GraphRAG required between 26% and 97% fewer tokens than alternative approaches, making it not just better at providing answers, but also cheaper and more scalable⁷.

Digging deeper into the topic of accuracy, it’s not just whether an answer is correct that’s important; it’s also how useful the answers are. What people have been finding with GraphRAG is that not only are the answers more accurate, but they are also richer, more complete, and more useful. LinkedIn’s recent paper describing the impact of GraphRAG on their customer service application provides an excellent example of this. GraphRAG improves both correctness and richness (and therefore usefulness) for answering customer service questions, reducing median per-issue resolution time by 28.6% for their customer service team⁸.

A similar example comes from a GenAI workshop taught by Neo4j and with our partners at GCP, AWS, and Microsoft. The sample query below, which targets a collection of SEC filings, provides a good illustration of the kinds of answers that are possible when using vector + GraphRAG vs. those that one obtains when using vector-only RAG:

Note the difference between describing the characteristics of companies likely to be impacted by a lithium shortage, and listing specific companies that are likely to be. If you are an investor looking to rebalance your portfolio in the face of a change in the market or a company looking to rebalance its supply chain in the face of a natural disaster, having access to the latter and not just the former can be game changing. Here, both answers are accurate. The second one is clearly more useful.

Episode 23 of Going Meta by Jesus Barrasa provides another great example using a legal documents use case, starting with the lexical graph.

Those observing the X-sphere and who are active on LinkedIn will spot new examples coming out regularly from not just the lab but the field. Here, Charles Borderie at Lettria gives an example of vector-only RAG contrasted with GraphRAG, against an LLM-based text-to-graph pipeline that ingests 10,000 financial articles into a knowledge graph:

As you can see, not only did the quality of the answer improve markedly with GraphRAG vs. plain RAG, but the answer took one-third fewer tokens.

One last notable example I will include comes from Writer. They recently announced a RAG Benchmarking Report based on the RobustQA framework, comparing their GraphRAG-based approach⁹ to competitive best-in-class tools. GraphRAG resulted in a score of 86%, which is a significant improvement from the competition, whose scores ranged between 33% and 76%, with equivalent or better latency.

Every week I meet with customers across many industries who are experiencing similar positive effects with a wide variety of GenAI applications. Knowledge graphs are unblocking the path for GenAI by making the results more accurate and more useful.

#2: Improved Data Understanding, Faster Iteration

Knowledge graphs are intuitive both conceptually and visually. Being able to explore them often reveals new insights. An unexpected side benefit that many users are reporting is that once they’ve invested in creating their knowledge graph, they find that it helps them build and debug their GenAI applications in unexpected ways. This has partly to do with how seeing one’s data as a graph paints a living picture of the data underlying the application. The graph also gives you hooks for tracing answers back to data, and tracing that data up the causal chain.

Let’s look at an example using the lithium exposure question above. If you visualize the vectors, you will get something like this, except with far more rows and columns:

When you work with your data as a graph, you can apprehend it in a way that’s just not possible with a vector representation.

Here is an example from a recent webinar from LlamaIndex¹⁰, showing off their ability to extract the graph of vectorized chunks (the lexical graph) and LLM-extracted entities (the domain graph) and tie the two together with “MENTIONS” relationships:

(You can find similar examples with Langchain, Haystack, SpringAI, and more.)

Looking at this diagram, you can probably start to see how having a rich structure where your data resides opens up a wide range of new development and debugging possibilities. The individual pieces of data retain their value, and the structure itself stores and conveys additional meaning, which you can use to add more intelligence to your application.

It’s not just the visualization. It’s also the effect of having your data structured in a way that conveys and stores meaning. Here is the reaction of a developer from a well-known fintech a week into introducing knowledge graphs into their RAG workflow:

This developer’s reaction aligns well with the test-driven development assumption of verifying – not trusting – that answers are correct. Speaking for myself, I get the heebie-jeebies handing 100% of my autonomy over to SkyNet to make decisions that are entirely opaque! More concretely though, even AI non-doomers can appreciate the value of being able to see that a chunk or a document tied to “Apple, Inc.” should really not be mapped to “Apple Corps”. Since the data is ultimately what’s driving GenAI decisions, having facilities at hand to assess and assure correctness is all but paramount.

#3: Governance: Explainability, Security, and More

The higher the impact¹¹ of a GenAI decision, the more you need to be able to convince the person who will ultimately be accountable if it goes wrong to trust the decision. This typically involves being able to audit each decision. It also requires a solid and repeatable track record of good decisions. But that isn’t enough. You also need to be able to explain the underlying reasoning to that person when they call a decision to the mat.

LLMs don’t offer a good way of doing this on their own. Yes, you can get references to the documents used to make the decision. But those don’t explain the decision itself – not to mention the fact that LLMs are known to make up those references! Knowledge graphs operate at an entirely different level, making the reasoning logic inside of GenAI pipelines much clearer, and the inputs a lot more explainable.

Let’s continue with one of the examples above, where Charles from Lettria loads up a knowledge graph with extracted entities from 10,000 financial articles and uses this with an LLM to carry out GraphRAG. We saw how this provides better answers. Let’s get a look at the data:

Seeing the data as a graph is the first part. The data is also navigable and queryable and can be corrected and updated as time goes on. The governance advantage is that it becomes far easier to view and audit the “world model” of the data. Using a graph makes it more likely that the responsible human who is ultimately accountable for the decision will understand it, relative to being served up the vector version of the same data. On the quality assurance side, having the data in a knowledge graph makes it a lot easier to pick out errors and surprises in the data (pleasant or otherwise), and trace them back to their source. You can also capture provenance and confidence information in the graph and use this not just in your calculation but your explanation. This just isn’t possible when you’re looking at the vector-only version of the same data, which as we discussed earlier is pretty inscrutable to the average – and even above-average!–human.

Knowledge graphs can also significantly enhance security and privacy. This tends to be less top of mind when building a prototype, but it’s a critical part of the path to production. If you’re in a regulated business such as banking or healthcare, the access any given employee has to information probably depends on that person’s role. Neither LLMs nor vector databases have a good way of limiting the scope of information to match up with the role. You can readily handle this with permissions inside a knowledge graph, where any given actor’s ability to access data is governed by the database, and exclude results that they aren’t allowed to see. Here is a mock-up of a simple security policy that you can implement in a knowledge graph with fine-grained access controls:

Knowledge Graph Creation

People often ask me what it takes to build a knowledge graph. The first step in understanding the answer is to know the two kinds of graphs most relevant to GenAI applications:

1. The Domain graph is a graph representation of the world model relevant to your application. Here is a simple example:
   
2. The Lexical graph¹² is a graph of document structure. The most basic lexical graph has a node for each chunk of text:
   

People often expand this to include relationships between chunks and document objects (such as tables), chapters, sections, page numbers, document name/ID, collections, sources, and so on. You can also combine domain and lexical graphs like so:

Creating a lexical graph is easy and largely a matter of simple parsing and chunking strategies¹³. As for the domain graph, there are a few different paths depending on whether the data you’re bringing in comes from a structured source, from unstructured text, or both. Luckily, tooling for creating knowledge graphs from unstructured data sources is rapidly improving. For example, the new Neo4j Knowledge Graph Builder takes PDF documents, web pages, YouTube clips, or Wikipedia articles, and automatically creates a knowledge graph from them. It’s as easy as clicking a few buttons, and lets you visualize (and of course query) both domain and lexical graphs of your input text. It’s powerful and fun, and significantly reduces the barrier to creating a knowledge graph.

Data about customers, products, geographies, etc. probably lives somewhere in your enterprise in a structured form, and can be sourced directly from wherever it lives. Taking the most common case where it’s in a relational database, you can use standard tools¹⁴ that follow tried-and-true rules for relational-to-graph mapping.

Working with Knowledge Graphs

Once you have a knowledge graph, there is a growing abundance of frameworks for doing GraphRAG, including LlamaIndex Property Graph Index, Langchain’s Neo4j integration as well as Haystack’s and others. This space is moving fast, but we’re now at the point where programmatic methods are becoming straightforward.

The same is true on the graph construction front, with tools such as the Neo4j Importer, which has a graphical UI for mapping & importing tabular data into a graph, and Neo4j’s new v1 LLM Knowledge Graph Builder mentioned above. The picture below summarizes the steps for building a knowledge graph.

The other thing you’ll find yourself doing with knowledge graphs is mapping human-language questions to graph database queries. A new open source tool from Neo4j, NeoConverse, is designed to help with natural language querying of graphs. It’s a first solid step forward toward generalizing this¹⁵.

While it’s certainly the case that graphs require some work and learning to get started with, there is also good news in that it’s getting easier & easier as the tools improve.

Conclusion: GraphRAG is the Next Natural Step for RAG

The word-based computations and language skills inherent in LLMs and vector-based RAG offer good results. To get a consistently great result, one needs to go beyond strings and capture the world model in addition to the word model. In the same way that Google discovered that to master search, they needed to go beyond mere textual analysis and map out the underlying things underneath the strings, we are beginning to see the same pattern emerge in the world of AI. This pattern is GraphRAG.

Progress happens in S-curves: as one technology tops out, another spurs progress and leapfrogs this prior. As GenAI progresses, for uses where answer quality is essential; or where an internal, external, or regulatory stakeholder requires explainability; or where fine-grained controls over access to data for privacy and security is needed, then there’s a good chance your next GenAI application will be using a knowledge graph.

You Can Experience GraphRAG Firsthand!

If you’re ready to take the next step with GraphRAG, I invite you to try the Neo4j LLM Knowledge Graph Builder. This simple web app lets you create a knowledge graph in just a few clicks, from unstructured text sources like PDFs, web pages, and YouTube videos. It’s the perfect playground for experiencing the power of GraphRAG firsthand.

With the LLM Knowledge Graph Builder, you can:

• Connect to your free cloud-based Neo4j instance and build a graph from your favorite text sources.
• Explore your newly created knowledge graph with interactive visualizations.
• Chat with your data and put GraphRAG to the test.
• Integrate your knowledge graph into applications and unlock new insights.

To get started, spin up a free AuraDB instance and build your knowledge graph. You can learn more about the Neo4j LLM Knowledge Graph Builder and get a guided tour here!

Acknowledgments

A great many people contributed to this post. I’d like to acknowledge all of you who share your learnings, writings, and code—many examples of which are cited here—and encourage you to keep doing so. It is by sharing as a community that we all learn.

I would also like to thank the many people who see the importance of GraphRAG and who generously offered their time to review and comment on the post itself. In many cases, this was informed by examples showing up in their world.

Rather than attempting to name everyone, I’d like to call out some of the people outside of what you would normally think about as the “graph world.” We are together seeing GraphRAG as not only an important trend but as a convergence between two worlds.

Having said all of this, my deepest thanks to all of you, including (alphabetically by last name):

• Harrison Chase, CEO of Langchain
• Ali Ghodsi, CEO of Databricks
• Rod Johnson, Investor and Founder of SpringSource
• Douwe Kiela, CEO of ContextualAI and Co-inventor of RAG
• Christina Li, FPV Ventures
• Jerry Liu, CEO of LlamaIndex
• Owen Robertson, Principal, DTS
• Milos Rusic, CEO of deepset / Haystack

Supplement: Further Reading

There’s been a lot written about this topic, with new insights and examples appearing every day. While I can’t hope to provide a comprehensive list, here are a few particularly good pieces you can check out if you’re interested in learning more:

Since 2016, disinformation campaigns have steadily eroded Americans’ faith in democracy, and they pose an unprecedented threat to the 2024 U.S. presidential election, according to political scientists and researchers. Bad actors have become particularly adept at leveraging fake social media advertising accounts to overwhelm users with political propaganda.

Fortunately, graph databases like Neo4j are built to fight this kind of disinformation. Neo4j enables social media platforms and election researchers to identify malicious accounts by uncovering key connections between them—connections that would otherwise go undetected.

From One Set of Credentials, Many Shared Accounts

Every social media advertising account is identified by a set of credentials: phone number, email address, website address, etc. In a Neo4j graph database, these accounts and credentials are represented as nodes, which are connected by directional relationships:

It’s not uncommon for multiple accounts to share the same credentials. A business owner, for example, might want to create separate accounts for different product lines but link them all to one email address for convenience. Similarly, family members living in the same household might create individual accounts but use one phone number.

The situation becomes more complex, however, when bad actors create fake accounts to spread disinformation, engage in coordinated inauthentic behavior, or conduct merchandise spamming. In those cases, shared credentials become red flags that require further investigation.

This is where graph databases excel—they allow us to unravel intricate webs of connections between accounts and identify hidden patterns that other database technologies would miss.

Identifying Connections Between Accounts to Unmask Bad Actors

Graph databases are uniquely suited to handle the problem of identity resolution—figuring out whether multiple records or entities in a dataset actually reference the same real-world thing.

In a Neo4j graph, we see exactly how individual accounts relate to credentials, and we can look for discrete areas of the graph—subgraphs —in which all the nodes are connected by their relationships to given credentials, regardless of the relationship direction. This kind of subgraph, whose nodes are all reachable from one another, is known as a component.

The accounts in a component are likely to be managed by the same individual or organization. For example, consider a scenario in which three accounts share the same email address:

Account A (blue) is John’s personal account, account B (green) is John’s business page for his coffee shop, and account C (yellow) is a suspicious account promoting counterfeit merchandise.

Using a graph database, we represent these accounts as nodes and establish relationships between them and their individual emails, which in this case happen to be shared email addresses. By analyzing the subgraph that emerges, we can infer that Accounts A and B are both managed by John and that John’s activities warrant further investigation due to the suspicious nature of Account C.

Distinguishing Between Legitimate and Malicious Connected Accounts

While shared credentials can raise suspicions, it’s crucial to recognize when connected accounts are legitimate. There are many valid reasons for managing multiple accounts. A social media manager, for example, might handle several pages for different clients, all linked to their professional email addresses. Alternatively, a local political party may have support accounts for various candidates.

Graph databases allow us to analyze patterns and anomalies within the weakly connected components to distinguish between legitimate and malicious accounts. One red flag is a sudden surge in new accounts. If someone creates many new accounts with the same credentials in a short period of time, they may be doing so maliciously. This kind of suspicious weakly connected component is on the right:

By analyzing graph patterns within components, we can identify and flag suspicious accounts for additional analysis, while ensuring that legitimate accounts can continue to operate and advertise effectively.

Analyzing Deleted Accounts to Determine Their Legitimacy

Another challenge in identity resolution arises when accounts are deleted, either by the company because the accounts violated platform policies or by the users themselves. Deleted accounts leave behind a trail of relationships that can provide insights into their legitimacy.

Graph databases maintain the historical data and relationships associated with deleted accounts, enabling us to track and analyze them. Closely examining deleted accounts and their associations allows us to make inferences about their legitimacy.

For example, if a deleted account had a history of authentic user engagement and legitimate advertising activities, it is more likely to have been a genuine account. Conversely, suspicious behavior, lack of engagement, and connections to other flagged accounts may indicate that a deleted account was created for malicious purposes.

By leveraging graph databases to analyze deleted accounts and their relationships, we can gain a more comprehensive understanding of the overall identity resolution puzzle, even when pieces go missing.

Connecting Accounts Without Common Credentials

Things get really interesting when we start to find connections between accounts that don’t share credentials. In the illustration below, we can see that the two green nodes, Account B and Account C, share an email address and phone number. But the graph also reveals something that would be extremely difficult to discern in tabular data: Account A (blue) and Account E (yellow) belong to the same network of accounts, even though they don’t share any credentials.

Ensuring Integrity in Political Advertising With Relationship Analysis

Identity resolution through relationship analysis is critical to maintaining the integrity of social media advertising during presidential elections in the U.S. By leveraging graph databases to identify and examine weakly connected components, we can peel back layers of complexity and distinguish between legitimate and malicious accounts.

Once we’ve mapped out malicious account networks, revealing influence previously hidden in thickets of account credentials, we can begin to rein in malicious activity, from coordinated inauthentic behavior and account masking to merchandise spamming.

Connected data reveals new ways to solve pressing business problems, even as data sizes increase. With a knowledge graph, firms can unlock the full potential of data assets for AI applications. This year’s FIMA survey shows that 50% of data executives are exploring or planning to use knowledge graphs for AI, and another 20% reported interest.

Firms have achieved early wins with AI to enhance the customer experience, internal processes, personal productivity, and cost-saving efforts. However, the true value of AI lies in the ability to uncover deep meaning in data, enabling us to make more strategic decisions where precision and accuracy matter. That’s where knowledge graphs come in.

A Knowledge Graph Primer

Knowledge graphs (KGs) underpin a wide range of applications, from drug discovery and consumer-facing personalization systems to critical infrastructure like public transport, power grids, and supply chain management.

A KG provides a powerful framework for organizing and linking data of all types. KGs capture relationships between entities from a real-world domain by using a graph structure – which formats data as an interconnected network – rather than a table with rows and columns.

Example of a knowledge graph.

For example, in cybersecurity, the KG acts as a data representation of an IT infrastructure, with all its security tools, applications, and dependencies. By modeling their IT network in a graph data model, security teams gain a holistic understanding of the system and can apply predictive analytics to find their own vulnerabilities before attackers do.

Over the last 15 years, Neo4j has helped companies use KGs to reason about complex datasets. Because KGs excel at connecting data across disparate systems and data sources, they work well for applications requiring real-time analytics. For example, in fraud detection, KGs show real-time behavioral patterns in the data that indicate fraudulent activities. For one Fortune 500 company, using a KG saved millions of dollars in fraud prevention each year while cutting manual review time in half. Another Neo4j customer boosted fraud detection rates by 200%.

The GenAI Enabler

The arrival of generative AI brought a demand for AI models to produce better, more accurate responses. KGs act as a control for Large Language Models (LLMs) by enabling knowledge-based reasoning based on the connections in the data – ultimately leading to AI we can trust. 48% of FIMA survey respondents plan to use techniques that aim to improve responses generated by AI. It’s no surprise that a combined 70% of respondents stated that they’re planning to use or are interested in exploring knowledge graphs this year.

Last month, Microsoft released a study where researchers combined a knowledge graph with an LLM, an approach termed GraphRAG. The results demonstrated increased comprehensiveness, explainability, and diversity of viewpoints in the LLM’s responses.

Deloitte has recognized KGs as the industry standard for enterprise-ready AI, and many leading analyst firms agree. According to Gartner, “GenAI models are being used in conjunction with knowledge graphs to deliver trusted and verified facts to their outputs, as well as provide rules to contain the model” (Emerging Tech Impact Radar: 2023).

Knowledge Graphs for the Enterprise

Knowledge graphs power a new generation of AI that is uniquely capable of providing a deep understanding of data. When evaluating a knowledge graph offering, use these criteria to determine enterprise readiness:

• Traverses multi-degree relationships with millisecond latency.
• Scales for the enterprise with transaction support, high availability, granular security controls, and full ACID compliance.
• Delivers an integrated set of tools for data loading, visualization, and analysis.
• Supports AI development with enablers such as GraphRAG, free-text indexing, graph algorithms, and ML pipelines.
• Integrates with open-source RAG frameworks and all major cloud providers.

Neo4j, the Graph Database & Analytics leader, helps organizations find hidden relationships and patterns across billions of data connections deeply, easily, and quickly. By capturing the connections between data as they exist in the real world, the Neo4j knowledge graph provides the data context required for advanced AI applications, including generative AI use cases. Neo4j supports LLM applications by combining the power of inferential reasoning with semantic similarity from native vector search.

Generative AI workflow with Neo4j knowledge graph.

Read The Artificial Intelligence Report by WBR Insights to learn more about building GenAI apps for the enterprise.

This article, “How GenAI Provides Deep Understanding With Knowledge Graphs,” was originally published in The Artificial Intelligence Report by FIMA in partnership with WBR Insights, Cognaize, Neo4j, and Redis.

In a hyper-connected, data-rich world, enterprises need to understand complex relationships within large, diverse datasets. Customer interactions, for example, involve tracking behavior, preferences, and patterns across online platforms, physical stores, and social media. Organizations must understand all these relationships to optimize strategies and improve the customer experience.

Today marks the introduction of a validated partner solution between Databricks and Neo4j. This connector will allow our joint customers to seamlessly combine structured and unstructured data, discover hidden patterns across billions of data connections, enhance contextual understanding within their data, and rapidly deliver enterprise-grade GenAI applications.

At Neo4j, we help organizations efficiently analyze the relationships within highly connected business data, even as data volumes grow. Neo4j use cases include fraud detection, supply chain and logistics, energy solutions, customer 360, and more. Developers using Neo4j with Databricks can now:

• Enhance analytics by ingesting data from Databricks to Neo4j. Create a seamless workflow to continuously process, analyze, and update data across both platforms, enabling real-time insights and decision-making.
• Uncover hidden patterns to generate deeper insights. Use Neo4j’s built-in graph algorithms and Cypher query language to uncover hidden patterns and deeper insights in data. Within Databricks notebooks, Neo4j Bloom and the Neo4j Visualization Library (NVL) can be used to explore data visually.
• Combine Neo4j knowledge graphs with graph retrieval-augmented generation (GraphRAG). Neo4j knowledge graphs improve RAG, resolving the critical issues of accuracy, explainability, and transparency – and unlocking GenAI’s full potential.

Here’s a closer look at how using Neo4j and Databricks together can deliver groundbreaking analytics and GenAI outcomes.

Enhancing Analytics by Ingesting Data From Databricks to Neo4j

The Neo4j Connector for Databricks seamlessly transfers data to Neo4j for analysis in a graph structure. The Neo4j Graph Database excels at handling interconnected data, making it an ideal platform for analyzing complex relationships and patterns. The connector can be used to read data from and write data to Delta tables from a Databricks notebook.

A code snippet that shows the options provided by the Connector with Databricks to ingest data into Neo4j to create nodes, labels, properties, and relationships.

Delta Lake and Neo4j are both ACID-compliant systems, which means they ensure data consistency, reliability, and integrity throughout the data pipeline.

For example, Delta tables are part of Delta Lake, an open-source storage layer that brings ACID transactions to Apache Spark and big data workloads. Neo4j and Delta Lake efficiently handle complex queries and real-time insights, enabling scalability by handling petabytes of data, which ensures data consistency and reliability through ACID transactions and optimizes data reads and writes with features like indexing and caching.

We built our Databricks connector with data availability and security in mind. Neo4j Aura has a 99.95% uptime SLA for real-time applications and complies with industry-standard regulations such as ISO 27001, GDPR, CCPA, SOC2, and HIPAA. To ensure that only approved data is analyzed, we are supported by Databricks Unity Catalog, as our connector can access the Databricks data layer through Databricks’ access control mechanisms. We also integrate with SSO providers like Microsoft Azure AD and Okta, offer encryption at rest through customer-managed keys (CMK), and employ role-based access control (RBAC) to safeguard access.

Uncovering Hidden Patterns in Data to Generate Deeper Insights

Neo4j Graph Database offers a developer-friendly schema that enables easy prototyping and evolution of data models from development to production. The property graph model allows storing properties directly within the graph, providing a powerful yet simple approach for architects to design and build graph models. This makes it easy to conceptualize and transition from whiteboard designs to actual implementations. A graph built on a Neo4j graph database combines transactional data, organizational data, and vector embeddings in a single database, simplifying the overall application design.

A native graph database allows users to quickly traverse through connections in their data, without the overhead of performing joins and with index lookups for each move across a relationship or other join strategies. We call this capability index-free adjacency – each node directly references its adjacent (neighboring) nodes, so accessing relationships and related data involves a simple memory pointer lookup. This makes native graph processing time proportional to the amount of data processed – it doesn’t increase exponentially with the number of relationships traversed and hops navigated.

Developers can also use pre-built graph algorithms and the Cypher query language to find patterns in data. Centrality, pathfinding, similarity, and many other Neo4j algorithms are useful for recommendation engines, supply chain optimization, identity and access management, and network monitoring.

Key use cases for Neo4j Graph Algorithms

Unlock the Potential of GenAI With Knowledge Graphs and Retrieval-Augmented Generation (RAG)

It’s hard to overstate the importance of knowledge graphs in GenAI development. Gartner considers knowledge graphs essential to the development of GenAI, and has urged data leaders to “leverage the power of LLMs with the robustness of knowledge graphs to build fault-tolerant AI applications.”

As the widespread use of GenAI has driven demand for better responses, knowledge graphs have excelled at improving LLM accuracy, relevance, and transparency. Knowledge graphs ground LLMs by representing relationships within data – which contextualizes responses – and by integrating both structured and unstructured data.

Through a technique called GraphRAG, LLMs retrieve relevant information from a knowledge graph using vector and semantic search and then augment their responses with the contextual data in the knowledge graph. Microsoft researchers have found that LLMs using GraphRAG not only deliver more comprehensive and explainable answers but also a greater diversity of viewpoints.

Developers using Databricks can accelerate GenAI app development by seamlessly incorporating GraphRAG capabilities into their projects. A variety of integrations make it easy to access popular AI frameworks and tools like LangChain and LlamaIndex.

GraphRAG for GenAI applications

Adding Critical Analytics and GenAI Capabilities to the Databricks Experience

Extracting insights from densely interconnected datasets and accelerating GenAI development are critical priorities for the modern enterprise. Our new integration with Databricks is specifically designed to help organizations meet these challenges – and to stay ahead of the GenAI and data analytics curve for years to come.

To get started with Neo4j on Databricks, read the quickstart documentation and get your projects started on Aura.

We’re thrilled to announce a new integration with Snowflake that brings the full spectrum of Neo4j’s industry-leading Graph Data Science (GDS) capabilities into the Snowflake AI Data Cloud for advanced AI insights and predictive analytics.

Snowflake end users seeking graph-enabled insights will now find it simple and straightforward to use GDS within the AI Data Cloud. Instead of moving data between Snowflake and Neo4j, end users can now perform graph analytics directly in the AI Data Cloud using Snowflake SQL – no graph expertise required.

Transforming Analytics in the AI Data Cloud With Graph Data Science

Neo4j’s Graph Data Science (GDS) brings a robust library of graph algorithms directly to Snowflake. Now, Snowflake users can initiate a fully serverless, isolated, and flexible graph analytics environment directly from SQL using Snowpark Container Services.

The dynamic nature of GDS on Snowflake allows for precise resource allocation and the option to power down when analytics processing concludes. The graph analysis results can be written back to the underlying Snowflake instance, seamlessly integrating with other data warehouse tables.

The new integration eliminates the need for complex and expensive data migration jobs between separate ML or database platforms outside Snowflake. End users will have a complete graph analytics offering in Snowflake that doesn’t require navigation between multiple environments. This streamlined, zero-ETL approach dramatically reduces time to value, allowing organizations to get more analytics projects into production faster, with tooling they already know.

Instant Graph Analytics With Robust Graph Algorithms

Neo4j brings the industry’s most extensive library of graph algorithms to Snowflake. Neo4j’s GDS algorithms – including similarity, pathfinding, community detection, and many others – open up a world of possibilities for advanced insights in generative AI, fraud detection, predictive analytics, feature development, recommendation engines, logistics, and more.

Graph Features for Machine Learning Pipelines

Neo4j’s graph algorithms can be used to generate predictive features to boost machine learning model performance, particularly where relationships between data points carry predictive signals. This is common for tasks related to fraud detection, entity resolution, recommendation systems, supply chain optimization, and customer journey, just to name a few. For example, statistics around centrality and clusters in payment and identity networks can boost performance for credit and insurance claim fraud detection models.

Graph Vectors for GenAI Applications

Joint customers can enhance generative AI (GenAI) applications by creating graph vectors that improve the relevance, accuracy, and personalization of large language model (LLM) responses. Neo4j GDS enables you to generate vectors using graph embedding algorithms that leverage structured, unstructured, and relationship data from a knowledge graph, significantly improving the grounding information for LLMs. Graph vectors can be used in various GenAI applications across industries such as banking, healthcare, supply chain, and more, providing domain-specific context and enhancing overall AI effectiveness. These features are part of a comprehensive GenAI stack within Snowflake, which includes both vector search and Snowflake Arctic LLM models.

Running Your First Algorithm

Once a Snowflake administrator has enabled GDS for Snowflake as a service, it’s easy to run your first algorithm. The first step is to create a GDS graph, an in-memory structure optimized for graph analysis that contains nodes connected by relationships. Both nodes and relationships can hold numerical properties.

Here’s an example of creating a graph and running the PageRank algorithm using Snowflake SQL.

First, project a nodes-and-relationship table into an in-memory graph structure that is optimized for graph analytics:

SELECT gds.graph_project('graph_name’,  ‘name_of_nodes_table’,  'name_of_relationship_table’);

Now you can perform graph algorithms on this in-memory structure:

SELECT gds.pagerank('graph_name’, 'pagerank_score');

After you’ve completed your graph computations, you can write the results stored in the graph back to a regular Snowflake table:

SELECT gds.write_nodeproperty('graph_name’, 'pagerank_score’, ‘result_table_name’);

The Limitless Potential of Graph in the AI Data Cloud

This integration enables our joint Neo4j and Snowflake customers to embrace the limitless possibilities of graph within Snowflake’s large-scale AI Data Cloud. We’re redefining how organizations extract insights from connected data by meeting users where they are in the AI Data Cloud. Product and engineering teams can now develop, scale, and operate applications without operational burden – a pivotal step towards using relationships in data to produce deeper, more comprehensive insights at an unprecedented pace.

Neo4j GDS in the AI Data Cloud is available for preview and early access, with general availability later this year on Snowflake Marketplace. Register your interest for an early access preview here.

The generative AI market has grown by a staggering 3,600% year over year according to O’Reilly, fueling renewed demand for knowledge graphs.

The technology of choice for highly connected, heterogeneous data, knowledge graphs work well for grounding large language models (LLMs). In fact, independent research from data.world and Microsoft have highlighted the benefits of using knowledge graphs for RAG use cases.

Building and operating a knowledge graph brings a plethora of design decisions. This article compares two methods: RDF from the original 1990s Semantic Web research and the property graph model from the modern graph database.

What Is a Knowledge Graph?

A knowledge graph is a semantically rich data model for storing, organizing, and understanding connected entities. A knowledge graph contains three essential elements:

• Entities, which represent the data of the organization or domain area.
• Relationships, which show how the data entities interact with or relate to each other. Relationships provide context for the data.
• An organizing principle that captures meta-information about core concepts relevant to the business.

Put together, these elements create a self-describing data model that enhances the data’s fidelity and potential for reuse. The organizing principle serves as a meta-layer, acting as a contract between the data and its users.

Recent research has demonstrated that the organizing principle can operate on a range of complexity levels, from simple labels on nodes with named relationships to intricate product hierarchies (e.g., Product Line -> Product Category -> Product) to a comprehensive ontology that stores a business vocabulary. The key is to select the appropriate level of complexity for the job at hand.

What Is RDF?

RDF stands for Resource Description Framework, a W3C standard for data exchange on the Web. It is also often (mis)used to describe a particular approach to managing data.

As a framework for representing the Web, RDF captures structure using a triple, the basic unit in RDF. A triple is a statement with three elements: two nodes connected by an edge (also known as a relationship). Each triple is identified by a Uniform Resource Identifier (URI) as subject-predicate-object:

• The subject is a resource (node) in the graph;
• The predicate represents an edge (relationship); and
• The object is another node or a literal value.

Here’s an example of an RDF model that represents a brother and sister, Daniel and Sunita. Sunita owns and drives a Volvo she purchased on January 10, 2011. Dan has also driven Sunita’s car since March 15, 2013:

The RDF model became well-known around the same time as knowledge graphs under the umbrella of The Semantic Web. The goal of the Semantic Web was to add machine-readable data with well-defined semantics to the public web using W3C standards. This would allow software agents to treat the web as a vast distributed data structure. By combining metadata (such as ontologies) and query languages (such as SPARQL), these agents would be able to infer knowledge from the web’s data.

Sadly, this excellent vision never came to pass. The technologies developed under the umbrella of the Semantic Web found their way into the database world as “triple stores,” where the same techniques were applied to local databases (rather than the Web). However, SQL databases were dominant at the time, largely edging out triple stores.

With the rise of the graph database, triple stores began to market themselves as a type of graph database. Triple stores typically use SPARQL as their query language. As with the grand vision of the queryable Web, SPARQL allows users to reason over sets of triples in the database, guided by the rules defined in an ontology. However, the RDF approach has significant drawbacks.

The Practical Challenges of Implementing RDF

Implementing a knowledge graph with RDF can be challenging because RDF wasn’t designed with database systems in mind. For one, it’s not possible to identify unique relationships of the same type between two nodes in RDF. When a pair of nodes is connected by multiple relationships of the same type, they are represented by a single RDF triple. The expressiveness of the data model is limited since you can’t capture scenarios where multiple distinct relationships of the same type exist.

Because many-to-many relationships cannot be modeled in RDF, users are forced to build workarounds that make the model more complex without adding value. In a social network, for example, users typically follow many other users and, in turn, are followed by others. To store these multiple relationships, RDF must introduce more triples, with new nodes, to model the properties of each relationship. This workaround is analogous to using join tables in a relational database.

While ontologies make RDF data “smarter” in the sense that they advertise a processing model for that data, there’s no guarantee that a SPARQL query will ever terminate. This leaves the user waiting – potentially indefinitely – for an answer. Computability of queries isn’t the only downside, however, in fact ontology-driven approaches like RDF pose practical challenges for implementations:

1. Skill building. Acquiring the expertise to create standard ontologies takes considerable time and effort. This investment brings risks and inefficiencies, as the highly specialized nature of these skills limits their transferability across the organization.


2. Resource-heavy. Building an ontology is a time-consuming process, and the ontology must be complete to deploy the knowledge graph. This significantly delays time to business value.


3. Fidelity to the business reality. Ontologies perpetually lag behind the business domains they support because the business moves more quickly than ontology updates. As a result, the knowledge graph becomes a reflection of the business as it was in the past, not its current state.

There are situations when it makes sense to invest in an ontology approach like RDF. An ontology that bridges the gap between two local domains is useful when those domains are individually well understood and information exchange is the goal. But importantly, ontologies can also be used outside of the semantic web stack since they are just graphs, after all.

What Are Property Graphs?

Information is organized as nodes, relationships, and properties in a property graph. Nodes are tagged with one or more labels, identifying their role in the network. Nodes can also store any number of properties as key-value pairs.

Relationships provide directed, named connections between two nodes. Relationships always have a direction, a type, a start node, and an end node, and they can have properties, just like nodes. Although relationships are always directed, they can be navigated efficiently in either direction.

Here’s an example property graph model that again shows the same brother and sister (Daniel and Sunita) that we looked at earlier:

The property graph model was designed to be a flexible and efficient way to store connected data, allowing for fast querying and traversal. It follows the important principle of storing data in a format that closely resembles the logical model – like how we might sketch out the data model on a whiteboard.

The property graph model reflects the logical data model that has been created to represent a company’s data and its relationships. In a native property graph like Neo4j, the physical storage model is isomorphic to the logical model. What you draw is what you store.

Property graphs have their own ISO standard query language called GQL, with Cypher being the most widely used implementation. Cypher and GQL differ from SPARQL as they are declarative languages that focus on pattern matching and not reasoning. In most practical scenarios, pattern matching suffices, and queries written in these languages will finish executing in a time proportional to the size of the graph being explored.

The Advantages of Using a Property Graph

Building a knowledge graph with a property graph database is straightforward compared to the alternatives. Unlike RDF, property graphs were designed as a database model (rather than data exchange format) for applications and analytics. Property graphs easily handle many-to-many relationships or multiple relationships of the same type between the same two nodes. They also provide much greater flexibility throughout the development process because they aren’t limited to a predefined structure. A property graph makes it possible for the business to build, expand, and enrich the knowledge graph over time. The main pros of using a property graph include:

• Simplicity: Property graphs are simple and quick to set up and use. Knowledge graphs built with property graphs have low complexity for new and experienced users.
• Detailed: User data can easily be stored in both nodes and relationships, ensuring that entities and their connections accurately reflect the business domain.
• Interoperable: Neo4j can consume and produce RDF for interoperability with legacy triple stores and integrate with any modern database system, including relational and document stores.
• Standards Compliant: With ISO GQL, implementers can have the same confidence for graph implementations that they have for relational databases that use ISO SQL.

The ROI of Implementing a Property Graph

The results of using a property graph for knowledge graph implementation speak for themselves. From biotechnology to pharmaceuticals to space exploration, organizations in different industries solve complex problems, reduce costs, and achieve breakthroughs thanks to the flexibility and scalability of the property graph model.

NASA: Unlocking Decades of Project Data for Faster, Smarter Space Exploration

Though NASA has project data going back to the late 1950s, organizational silos kept valuable insights hidden until a knowledge graph was implemented.

NASA converted its database into a property graph, which now stores millions of historical documents. The property graph database allows engineers to access information about past projects, enabling engineers to identify trends, prevent disasters, and incorporate lessons learned into new projects.

The property graph strategy has already saved NASA millions and years of research and development towards their Mission to Mars planning.

Basecamp Research: Mapping Earth’s Biodiversity for Biotechnological Breakthroughs

Basecamp Research built the world’s largest knowledge graph of Earth’s natural biodiversity using Neo4j’s property graph database. By collecting and connecting biological, chemical, and environmental data from across the globe, Basecamp has expanded known proteins by 50% and documented over 5 billion biological relationships.

This knowledge graph has become a superior resource for protein design applications and generative AI models, fostering the development of improved drugs, food products, and diagnostics.

Basecamp Research uses a knowledge graph to map Earth’s biodiversity, capturing proprietary protein and genome sequences and environmental and chemical data. The resulting knowledge graph, containing over 5 billion relationships, reveals intricate biological networks and has expanded known proteins by 50%.

Novo Nordisk: Streamlining Clinical Trials with a Connected Data Landscape

Novo Nordisk uses a Neo4j-based knowledge graph to power StudyBuilder, streamlining the process of clinical data collection and reporting in healthcare.

The property graph helps navigate the highly connected landscape of data standards (CDISC, SNOWMED, UCUM, etc.), allowing StudyBuilder to ensure end-to-end consistency, built-in compliance, automation, and content reuse. This innovative approach to handling study specifications has been shared with the pharmaceutical community as an open-source project.

Property Graph or RDF for My Knowledge Graph?

While graphs are powerful, not all graphs are equal. The property graph model offers the most advanced approach for analytics and application development. Despite this, the RDF model has kept a toehold in some industries since its semantic web heyday.

	Property Graph	RDF
Store as native graphs		◯
Efficient, low-friction modeling		◯
Ease of getting started and incremental changes		◯
Efficient query of entity->property and relationship->property		◯
Query Language	Cypher/GQL	SPARQL
Speed/performance		◑
Analyze and enrich with data science		◯
Store transactional data natively		◯
Store semantic data natively
Leverage pre-built ontologies

The model you choose will, to an extent, be defined by circumstance. When free to choose, most folks will pick the property graph model for its simplicity and agility. Due to its wide use, the property graph model enjoys a large ecosystem of tooling, literature, training, a large pool of professionals, and ISO Standards support. By contrast, the RDF community inhabits a world where up-front design of ontologies and standards intended for interoperability on the Web are repurposed as database tools.

Graphs have moved on considerably since the Semantic Web era and contemporary best practices draw from both schools of thought. It’s more practical to use property graphs by default and layer in the organizing principles from the RDF world (taxonomies, ontologies) when your system needs them, not as a technical prerequisite.

The GQL ISO standard has just landed (April 11, 2024), marking a historical moment for graph database languages and a huge milestone in what has already been a relatively long history of language development.

In a previous post, I briefly sketched how the Neo4j proprietary implementation of Cypher is becoming GQL compliant. However, the Cypher world is bigger than Neo4j, its original creator: most graph databases support Cypher via the openCypher project.

The post starts with a brief history of property graph languages to give context to the most recent graph database practitioners. It then describes a vision of the openCypher project’s future in what will soon become a GQL world.

The Origins of Cypher: A New Beginning

The Cypher language emerged in 2010, during the early halcyon days of NoSQL. The early part of this decade saw more and more languages emerge for querying graph databases. Nearly every new vendor entering the stage invented their own language. In fact Neo4j itself had at least three languages for quite some time.

Cypher was declarative, unlike most other graph database query languages at the time. It was modeled after SQL, where you describe an outcome and let the database do the work of finding the right results. Cypher also strove to reuse wherever possible and innovate only when necessary.

Most graph languages took the alternative approach, called imperative graph querying. With this method, developers had to spell out each step the database should take. While easier for vendors, this burdened users. Slowly but surely, Neo4j users upvoted Cypher with their keyboards. Even though the earliest versions were limited, we saw users choose Cypher whenever they could.

By 2015, Cypher had gained a lot of maturity and evolved for the better, thanks to real-world hard knocks and community feedback.

Yet as time progressed, the graph query languages kept coming—still none of them with anything close to Cypher’s success. If this kept up, the graph database space would continue to accumulate new languages, making it more and more confusing for users and for the budding ecosystem of graph tools, connectors, and consultancies.

At Neo4j, we realized that if we cared about solving this problem, we needed to give away Cypher.

openCypher 9: The Need for Convergence

In October 2015, Neo4j launched a new open initiative called openCypher. openCypher not only made the Cypher language available to the ecosystem (competitors included!). It also included documentation, tests, and code artifacts to help implementers incorporate Cypher into their products¹. Last but not least, it was run as a collaboration with fellow members of the graph database ecosystem, very much in keeping with Neo4j’s open source ethos. This started a new chapter in the graph database saga: one of convergence.

openCypher proved a huge success. More than a dozen graph databases support Cypher, many dozens of tools & connectors also support Cypher, and there are tens of thousands of projects using Cypher, with tens of thousands of certified Cypher professionals, many universities and online courses that include Cypher, and hundreds of thousands of developers who know and use Cypher.

It turns out there is another step one could take: go from a de facto standard to a de jure standard. This entails going to an official standards body with global standing, investing time, rigor, and diligence across multiple parties and many, many meetings and documents, and coming out the other end with an iron-clad and thoroughly vetted standard.

The Advent of GQL: Setting New Standards

In 2016, Neo4j started work towards that goal. We approached other vendors about collaborating on a formal standard, participated in a multi-vendor and academic research project to build a graph query language from scratch on paper, and eventually joined ISO. Momentum reached a crescendo in 2018, when, just ahead of a critical ISO vote, we polled the database community with an open letter to vendors, asking the community if we database vendors should work out our differences and settle on one language, rather than minting out new ones every few months. Not surprisingly, the answer was a resounding yes. Challenge accepted! The die was cast.

In 2019, the International Organization for Standardization (ISO) announced a new project to create a standard graph query language. They called it GQL for Graph Query Language. Since then, Neo4j and several other database vendors have been diligently working to define a standard language.

Fast forward to today, and GQL is finally here. The ISO committee has officially published GQL as the new international standard for graph query languages. The publication of this standard holds immense potential for the future of graph query languages.

Looking Ahead: openCypher Becomes a Road to GQL

GQL is changing the graph query language scene. In this exciting new world, we have decided to keep the openCypher project alive. The rest of this post will explain what we are going to do to reboot the openCypher project and why we are doing it.

openCypher was initially meant as a language specification project, with an open forum for discussions on new language features followed by specification and community votes. In the new GQL world, ISO takes most of that role. Anyone interested in discussing language features should join ISO or any forum such as LDBC and actively participate in GQL development.

The other original role of openCypher was to help language developers by supplying useful artifacts. The openCypher community is quite big, and it is very likely that most openCypher implementers are now thinking about GQL and how to get there. Here, openCypher can still help.

The basic idea is to use the openCypher project to help Cypher database and tooling vendors on their road to GQL. All openCypher implementers, and all their users, start the road to GQL from a similar starting point, which is a very good one, given the similarities between Cypher and GQL. We can walk the remainder of the path together. Over time, openCypher will become a GQL implementation.

What Happens Now?

Our plan does not yet dot all the i’s and cross all the t’s, but here are the highlights.

We will ‘freeze’ the current openCypher 9, which stays as it is.

openCypher will start publishing openCypher Improvements Proposals (CIPs) that introduce variations and extensions to openCypher to make it GQL compliant. Only features coming from the GQL standard will be considered for inclusion. The CIPs will provide an explanation and audit trail, provide a tie-in to the GQL spec, and introduce GQL features in a way that is as least disruptive as possible for Cypher users.

Since ISO has already discussed and vetted these features while creating the GQL standard, the openCypher processes will slim down considerably. From the perspective of openCypher, the work of language design will all happen inside of ISO, making the openCypher work about implementing and reflecting the standard within the language artifacts.

We will also start making new, versioned openCypher releases on a regular time cadence (to be defined, let’s assume six months). A release will collect all the CIPs published in that period. We are working on a release naming strategy, but ideally, the name will mention the GQL standard we are working towards (GQL:2024) and some other component to indicate the progressive steps towards it.

openCypher will continue supplying artifacts: in addition to the language specification (CIPs), it will include an updated grammar that incorporates the core GQL syntax elements introduced by the published CIPs. We intend to keep the SDK up to date, but this will be second to updating the specs and grammar. Any help with SDK work will be welcomed.

openCypher has fulfilled its initial purpose, serving as the basis for a graph database lingua franca across much of the industry. It is heartwarming to the team that has been invested in curating openCypher to think that now that GQL is finally here, openCypher can still have a different but useful role in ramping implementers and users onto GQL. Our dream is to see all openCypher implementations becoming GQL-conformant implementations, after which we will all be speaking GQL! Let’s make it happen.

To learn more, the following blogs and documents provide additional information about the GQL standard, Neo4j Cypher, and openCypher:

• ISO GQL: A Defining Moment in the History of Database Innovation
• ISO/IEC JTC 1 GQL Database Language
• GQL: The ISO Standard for Graphs Has Arrived
• GQL is Here: Your Cypher Queries in a GQL World

Got your eye on NODES 2024? The deadline to submit your talk is fast approaching. Developers and data scientists from all over the world are getting ready to share their latest graph-powered projects at this year’s free online conference. If you’d like to join them, here are a few tips to help you craft your presentation, even if you’re new to the game!

But first, some background. You can choose from three session formats: a 30-minute talk with Q&A, a 10-minute lightning talk, or a two-hour hands-on workshop. Your talk should fall into one of this year’s four tracks: Applications, AI, Data Science, and Graphs. Submit your graph story by June 15.

Now, with the call-for-papers deadline creeping ever closer, let’s run through seven tips to help refine your idea and write your abstract.

1. Know Your Audience

Got a concept in mind? Cool. Now, think about your audience. This is key because it shapes your whole session. Are they devs skilled in multiple languages? What industries are they from?

2. Outline Your Talk

Sketch out the main points you’ll cover. What problem are you solving? How does graph tech come into play? And what demos or insights will you share?

3. Start Strong

Kick off with a line that sets up the problem your audience faces. For example: “Every developer should know how to ____.”

4. Sell Your Session

In a couple of sentences, explain what people will gain from your talk. This is the heart of your abstract – it tells them your session is for them. Be clear: “You’ll learn ____ and ____.” And highlight what makes your session unique. If you’re live-coding or sharing lessons learned, mention it.

5. Show You’re Human

Add a personal touch to your abstract. Mention your name! For instance: “In this session, Jamie will show you … “

6. Nail Your Title

Toss around a few title ideas. Try straightforward ones, then mix in some fun options. Pick one that grabs attention.

7. Get Feedback

Ask peers to review your title and abstract. Adjust based on their feedback, then hit “Submit”!

New to Speaking? Try a Lightning Talk.

If it’s your first time speaking or talking about graph technology, consider a lightning talk. They’re short — just 10 minutes. Got a neat trick or solution using graphs? Talk about it!

Need inspiration? Check out our blog post, “10 Ideas to Inspire You to Present at NODES 2024.”

The deadline is June 15th — don’t miss it. Good luck!

Neo4j has achieved full compatibility with AWS PrivateLink, earning AWS PrivateLink Service Ready status. This means AWS has validated our technical work and verified that customers are successfully using the Neo4j Aura–PrivateLink integration.

AWS PrivateLink allows organizations to securely access AWS services within a virtual private cloud (VPC), giving them greater control over their data and mitigating the security risks associated with exposing data to the public internet.

In this post, we’ll walk through the setup of AWS PrivateLink with a Neo4j AuraDS Enterprise deployment running on AWS. Note that this integration is offered exclusively on the Enterprise tier of Neo4j Aura — it’s not available on Aura Free or Aura Pro.

Before we start, here’s a look at the architecture of the deployment:

And here are the six steps we’ll walk through:

1. Set Up a Neo4j Aura Instance
2. Set Up Neo4j Network Access Configuration and AWS Endpoint
3. Test the Network Configuration
4. Connect to Neo4j Aura Using the Python API
5. Connect to Neo4j Browser
6. Connect to Neo4j Bloom

1. Set Up a Neo4j Aura Instance

OK, let’s get started! First, you’ll need to set up an Aura instance. To do so, log in to the Neo4j Aura Enterprise console. If you don’t have Neo4j Aura Enterprise, you can purchase it through the AWS Marketplace.

Once the instance is deployed, you will see its status as “Running” in the console. Note that we have deployed an AuraDS instance. This includes the Graph Data Science (GDS) interface. We’ll use that later when we write a brief Python program to interact with the cluster.

Also, note that the public URL in the password file you download during deployment will not work once PrivateLink is enabled and public access is turned off. You’ll need to use the private URL instead.

2. Set Up Neo4j Network Access Configuration and AWS Endpoint

Now we can set up a new Network Access Configuration. This is an object in the Aura console. Click on “Network Access” under the security menu. Note that this menu will not be present in lower tiers of Aura or if users do not have admin access.

The configuration is both regional and specific to the product being used. In this case, we’re using AuraDS.

After we click through, the Aura console provides a nice walkthrough for PrivateLink setup. Once the automated configuration is complete, we’ll follow those steps.

Next, go into the AWS console and create a new endpoint in your VPC. Be sure to enable private DNS for the endpoint! You’ll need it to resolve the names of the Aura endpoints later.

In this case, we’re creating a security group with extremely open permissions. You might want to lock yours down a bit more.

Now the endpoint is up and running! Note that “Private DNS names enabled” says yes.

Now we paste the endpoint into the Aura console, and Aura makes a connection.

When it’s successful, we’ll see “Accepted.”

Now, here’s the key part. We’re going to turn off the public network.

Take note of the Neo4j knowledge base (KB) article on testing connectivity. We’ll come back to that in the next section.

On the Aura console, you’ll see some advice about how to use VPNs. For this walkthrough, we’re going to take a different approach and use a jump box instead.

When you dismiss the dialogs, public access might still show as “Enabled.” Give it a minute …

When done, public access will show “Disabled” and private access will show “Enabled.”

That’s it! We’ve set up PrivateLink access. Next, we’ll verify that it’s working properly.

3. Test the Network Configuration

First, let’s confirm that we’ve disabled public access by trying to connect to the old public address of your Aura instance. It should give you an error. This is one time when an error message is actually a good thing!

Now let’s spin up an EC2 and try to connect …

Using EC2 Instance Connect, let’s open a terminal to our new machine.

Per the KB article, let’s try nslookup. The command for our deployment is as follows. (Yours will have a different address.)

nslookup 51a2cfea.production-orch-0508.neo4j.io

Similarly, we can curl to hit the endpoint:

curl 51a2cfea.production-orch-0508.neo4j.io

According to the test KB article, we should now run netcat. AWS Linux doesn’t come with it installed, so we need to run the command:

sudo yum install nmap-ncat

Now we can run the commands:

nc -vz 51a2cfea.production-orch-0508.neo4j.io 443
nc -vz 51a2cfea.production-orch-0508.neo4j.io 7687
nc -vz 51a2cfea.production-orch-0508.neo4j.io 7474

The DNS seems to be working, and we’re able to resolve all the addresses we need.

4. Connect to Neo4j Aura Using the Python API

Now, let’s run a simple Python program to make sure that our EC2 instance can connect to the database.

First, we’ll need to install the driver, which is python-based. Of course, to grab that we need pip3 too. We can get all that with the commands:

sudo yum update && sudo yum install python3-pip
pip3 install --user graphdatascience

With that complete, run the command python3. We can then write a little program:

from graphdatascience import GraphDataScience

# Username is neo4j by default
NEO4J_USERNAME = 'neo4j'

# You will need to change these to match your credentials
NEO4J_URI = 'neo4j+s://51a2cfea.production-orch-0508.neo4j.io'
NEO4J_PASSWORD = 'C1HDzgf_XF78s6P5CSkJIpeVtFzDEnxrVkFH4Fzsl-U'

gds = GraphDataScience(
    NEO4J_URI,
    auth=(NEO4J_USERNAME, NEO4J_PASSWORD),
    aura_ds=True
)
gds.set_database('neo4j')

In our terminal, we should see the following:

It all works! Next, let’s try the UI.

5. Connect to Neo4j Browser

We can’t publicly connect to the Neo4j UI, Workspace, as we disabled access, but we should be able to connect to Neo4j Browser and Bloom over the private network. Let’s start with Browser.

Make a Windows box and then RDP to it and connect from there.

Select a standard Windows box and make sure RDP is enabled.

I’m on a Mac and haven’t used Windows in ages, so I needed to install RDP from here: https://apps.apple.com/us/app/microsoft-remote-desktop/id1295203466?mt=12

Grab the RDP file for your new machine.

We’ll end up connected directly to Neo4j Browser. You could enable Neo4j Workspace by opening a support ticket, but we’re not going to use Workspace, so make sure it’s disabled.

Now fire up the RDP file we downloaded in Remote Desktop. Open a web browser and connect to the database. Note that the private DNS does not populate correctly in the Connect URL. You’ll have to specify it manually yourself.

We can test the connectivity and DNS resolution on our Windows box using these commands:

Test-NetConnection 51a2cfea.production-orch-0508.neo4j.io -Port 7687
Test-NetConnection 51a2cfea.production-orch-0508.neo4j.io -Port 7474
Test-NetConnection 51a2cfea.production-orch-0508.neo4j.io -Port 443

6. Connect to Neo4j Bloom

Neo4j’s BI tool, Bloom, is specifically designed for graphs. To connect to Bloom, go to the Aura console. Because we disabled Workspace, you can see the “Explore” and “Query” buttons. Click “Explore” to open Bloom.

Copy the URL and paste it into your web browser within the RDP session.

Log in with our credentials.

Enter your password and boom — we’re in Bloom, connected securely over RDP and PrivateLink.

Private Deployment

I hope you found this post helpful and the setup straightforward. The deployment with AWS PrivateLink and Neo4j AuraDS Enterprise is entirely private—neither the database, browser, nor Bloom is ever publicly available.

We spent most of the post verifying that the setup was working properly. The approach we tried here with a jumpbox is a great way to get started, but another common approach is using a VPN to access the peered network.

If you have any questions about the setup process, please reach out to us at ecosystem@neo4j.com.

To learn more about how Neo4j and AWS work together, take a look at our AWS partnership page.

We’re excited to share that Neo4j has joined the Connect with Confluent technology partner program. This program helps businesses accelerate the development of real-time applications through fully managed integrations with Confluent Cloud. Thanks to the partner program, our customers now have the best experience for working with data streams directly within Neo4j, paving a faster path to powering next-generation customer experiences and business operations with real-time data. Keep reading to learn how the partnership expands the data streaming ecosystem.

Expanding the Data Streaming Ecosystem Together with Confluent

Connect with Confluent brings fully managed data streams directly to organizations through a single integration to the cloud-native and complete data streaming platform, Confluent Cloud. It’s now easier than ever for organizations to stream any data to or from Neo4j with a fully managed Apache Kafka® service that spans hybrid, multi-cloud, and on-premises environments. In addition, the program helps partners’ go-to-market efforts via access to Confluent engineering, sales, and marketing resources. This ensures customer success at every stage, from onboarding to technical support.

Build Real-Time Applications with Neo4j

“Joining the Connect with Confluent program marks an exciting milestone for Neo4j, as it underscores our commitment to empowering businesses with transformative data solutions, Matt Connon, vice president of indirect sales at Neo4j, expresses. “By partnering with Confluent, we’re poised to deliver greater value to our customers, enabling them to apply the combined power of graph technology and real-time event streaming. Together, we’ll unlock new insights, agility, and innovation, empowering organizations to thrive in an increasingly connected world.”

Integrating Neo4j with Confluent’s event streaming platform presents a powerful real-time data processing and analytics solution. By applying Neo4j’s ability to model and query highly connected data and Confluent’s distributed streaming architecture, businesses can unlock key use cases such as real-time recommendations, fraud detection, and network analysis.

For instance, organizations can stream data from various sources into Kafka topics, enrich and transform that data using Confluent’s portfolio of 120+ pre-built connectors, and then ingest it into Neo4j for graph-based analysis and insight generation. This integration enables businesses to detect patterns and relationships in data as they occur, leading to faster decision-making, enhanced customer experiences, and proactive risk management.

Additionally, by combining Neo4j’s graph algorithms with Confluent’s scalable infrastructure, companies can achieve unparalleled performance and scalability for their graph-based applications, ultimately driving innovation and competitive advantage in today’s data-driven landscape.

Learn how to configure the Confluent connector for Neo4j in this blog written by our very own Stu Moore, Senior Product Manager. We previously announced support for Confluent’s Custom Connectors to connect Confluent Cloud to AuraDB, our fully managed database as a service. In this blog, Stu shows you how easy it is to set up a sink connector to create or update data in a database in Aura when messages are written to a topic.

Get Started

Learn more about Neo4j Aura and try it for free at: neo4j.com/cloud/platform/aura-graph-database/.

Not yet a Confluent customer? Start your free trial of Confluent Cloud today. New sign-ups receive $400 to spend during their first 30 days — no credit card required.

Attention developers and data scientists! Do you have an exciting project or technique that involves graphs? If the answer is yes, we encourage you to present at NODES 2024, an online conference focused on graph-driven innovation.

It’s a great opportunity to share your knowledge, connect with others, and showcase your expertise.

Submit your proposals by June 15th and pick one of these talk tracks:

Applications: Libraries, Frameworks, and Platforms

Discover how developers use Neo4j to power inventive solutions across software stacks, cloud providers, and programming languages.

AI: Generative AI, Knowledge Graphs, and Retrieval-Augmented Generation

Explore the intersection of groundbreaking research and real-world applications using graph technologies and techniques.

Data Science: Machine Learning, Graph Data Science, and AI Models

Learn advanced techniques in data curation and maintenance designed to fuel AI models.

Graphs: Visualization, Data Integrations, and Tips & Tricks

Unlock the power of graphs, connect knowledge graphs to broader data systems, and uncover expert tips and tricks.

Need help crafting a presentation? Take a look at these notable talks from NODES 2023 for some inspiration:

1. Using LLMs to Convert Unstructured Data to Knowledge Graphs

Noah Mayerhofer, Software Engineer at Neo4j, generates knowledge graphs from unstructured data using LLMs for entity extraction, semantic relationship recognition, and context inference.

2. Create Graph Dashboards With LLM Powered Natural Language Queries

Niels de Jong, Consulting Engineer at Neo4j, uses LLM-powered queries in NeoDash to create Neo4j dashboards using text instead of Cypher.

3. Fine-Tuning an Open-Source LLM for Text-to-Cypher Translation

Enabling users to interact with Neo4j databases intuitively, Jonas Nolde, Machine Learning Engineer at berrybeat, fine-tunes an LLM to generate Cypher statements from natural language input.

4. Create Awesome Graph Visualizations From Your Data

Illustrating the awesomeness of graph visualizations, Sebastian Mueller, CTO of yWorks, explores different tools to generate impressive and meaningful results.

5. Build Apps with the New GenAI Stack from Docker, LangChain, Ollama, and Neo4j

Look inside the containers of the GenAI Stack from Docker, LangChain, Ollama, and Neo4j that add generative AI capabilities to your applications with Harrison Chase, CEO & Founder of LangChain, and Oskar Hane, Sr. Staff Software Engineer at Neo4j.

6. Knowledge Graph-Based Chatbot

Tomas Bratanic, a Graph & LLM enthusiast, demonstrates the use of a knowledge graph as a storage object to take control of the answers provided by the chatbot and avoid hallucinations.

7. Graph Machine Learning for 2024

Watch this forward-looking, technical overview of GML by Zach Blumenfeld, Product Specialist DS/ML at Neo4j, as he lays out a graph-centric AI architecture for predictive tasks.

8. Using Graph Theory to Model a Production Line and Predict Delivery Dates

Identifying bottlenecks and patterns in a production line, José Diogo Viana, Backend Engineer at Remote, demonstrates the usage of graph theory to improve production line processes.

9. NeoGenAI: Ontology Guided Loading of Business Data in Graph Database

Dattaraj Rao, Chief Data Scientist at Persistent Systems uses LLMs to load business data from disparate sources and schemas into graph databases based on a standard ontology.

10. Entity Resolution and Deduping: Best Practices From Neo4j’s Field Team

Mark Quinsland, Senior Field Engineer at Neo4j, describes common entity resolution and deduplication techniques for use cases in financial, health care, and others, with tools like GDS and Cypher.

Need more ideas? You can check out the complete NODES 2023 playlist on YouTube. Additionally, you can share your ideas with the community in the conference and event forum thread. You might even find someone interested in collaborating on a presentation.

The deadline to submit is June 15th. We’re excited to review your proposals!

Organizations of all sizes continually strive to improve data security, with enterprises often leading the way. At Neo4j, we take enterprise security seriously, which is why we’ve just released Customer Managed Keys (CMKs) for Neo4j Aura, our fully managed graph database as a service.

CMKs allow organizations to encrypt their Neo4j-managed graph databases with their own cloud-based keys, giving AuraDB and AuraDS Enterprise users an increased level of autonomy for security-focused operations.

Protect Data in Aura With Your Own Key

How do CMKs work in Aura? Let’s say you manage your cloud-based keys on AWS KMS. Now you can provide Aura with the Amazon Resource Name (ARN) of the key in the configuration panel without giving that information to any support personnel. We will recognize that as your Customer Managed Key, which will be used to encrypt data on Aura instances. Only you have control over your keys — including key policies, key rotation frequency, and key versions.

With CMKs, you can define access permissions and usage policies according to your specific security requirements. This ensures that only authorized users and services can access encrypted data.

Using CMKs will allow your organization to comply with strict data protection and privacy regulations — including GDPR, HIPAA, and PCI DSS — by implementing robust encryption and access controls.

Creating Customer Managed Key with Encryption Key ARN

After you’ve created a Customer Managed Key on Aura, you can start using it to encrypt data on new Aura or Graph Data Science instances. In the general availability release, you can add up to one key per region and product.

Get Started With Customer Managed Keys for Neo4j Aura

Customer Managed Keys brings additional enterprise-grade security features to the Aura platform. With control of essential compliance-focused functions in their own hands, AuraDB and AuraDS Enterprise users benefit from a new level of flexibility and security.

Customer Managed Keys is currently available for Amazon Web Services keys. We’re working hard to get Microsoft Azure and Google Cloud Platform keys available as Customer Managed Keys on Aura.

To get started, log in to the Aura Console or visit Neo4j Support.

Neo4j has empowered investigative journalists and other complex data uses for more than a decade. From the International Consortium of Journalists (ICIJ) using Neo4j to uncover the Panama Papers scandal in 2015 to NBC deploying Neo4j to expose Russian interference in the 2016 U.S. election, graphs are invaluable in analyzing mass data leaks and revealing hidden patterns in complex, connected data.

Now, Syracuse University’s Institute for Democracy, Journalism, and Citizenship (IDJC) is using Neo4j to shine a light on the murky world of political ad spending on social media. By using a Neo4j graph database to model the tangled network of political ads, funders, and candidates, the IDJC can uncover crucial insights into the flow of money and influence in our digital political landscape. In this post, we take a technical deep dive and explore how graph technology helps the IDJC surface insights that were previously hidden.

How Graph Reveals Hidden Connections in Political Ad Spending

So, what exactly is a graph database, and how does it differ from traditional data storage methods? In a graph database, each entity, whether a person, a political ad, or a funding organization, is represented as a “node.”

These nodes are then connected via “relationships.” For example, if a particular funder sponsors a political ad, a relationship would connect the “Funder” node to the “Ad” node. Similarly, if an ad mentions a specific candidate, a relationship would connect the “Ad” node to the “Candidate” node.

Instead of sifting through a massive, rigid data table to find relevant connections, researchers can follow the relationships between nodes to uncover hidden patterns and relationships. Graph Databases like Neo4j shine when we start looking for more nuanced patterns. For example, let’s consider the way an ad might “MENTIONS” a candidate. MENTIONS is the relationship in the graph between “Ad” and “Candidate”, as seen in the diagram below.

Is it supporting the candidate? Is it attacking the candidate? Adding this extra layer allows the relationship to be seen in a new light. Let’s take a look at an example of this new relationship. Here, we can see a funder that attacks a single candidate and supports another. Refining “MENTIONS” exposes an overview of the funder’s content strategy. While the relationship “FUNDED” still exists, a deeper understanding of the ad’s content informs the relationship.

One can imagine a pattern if we look at all of the funder’s ads. Perhaps many relationships are actually “SUPPORTED” and only occasionally “ATTACKED.” Equally possible is the opposite. A funder may be solely focused on attacking a candidate or, in the case of the primary, perhaps many. By using graphs, we can get a more holistic view of the funder’s actions by seeing the whole web of their messaging. Interesting stuff, but wait, there’s more! Let’s take this same concept and further expand past a single funder to understand the relationships of many funders.

The Power of Graph: A More Informed and Transparent Society

Where graphs can take you in this exploration is only limited by one’s imagination. Graphs can answer a variety of questions, such as: Are there certain funders or organizations that are disproportionately sponsoring ads containing misinformation or disinformation? Are there clusters or networks of ads and funders that are working together to spread misleading or false information? How do the relationships between funders, ads, and the spread of misinformation evolve over time? Important and compelling questions. This technology’s applications go far beyond political advertising. For example, graph databases have investigated complex financial crimes like money laundering and fraud. By analyzing the money flow between different accounts and entities, investigators can identify suspicious patterns and trace illegal activities back to their sources. Graph technology has also been used in healthcare and scientific research to uncover new insights and connections between variables and outcomes — and these are just a few examples in a world of endless possibilities.

As data volume and complexity continue to grow, graph technology will play a crucial role in helping us navigate and understand the intricacies of the information landscape. By embracing tools like graph databases and query languages, researchers and investigators can uncover hidden stories and insights that might otherwise go unnoticed —  ultimately helping build a more informed and transparent society.

The Far-Reaching Impact of the IDJC’s Graph-Powered Research

The IDJC’s pioneering research, powered by the cutting-edge capabilities of graph technology, has the potential to profoundly impact the integrity and transparency of our democratic processes. This work can empower voters to critically evaluate political messages and resist attempts to influence their opinions by exposing the hidden networks behind misleading or false information in social media ads. Furthermore, the tools and insights developed through this research can provide journalists and watchdog groups with powerful new means to investigate the sources and spread of misinformation and disinformation in political campaigns, holding bad actors accountable. The findings of the IDJC’s research can also inform policymakers and regulators as they consider new rules for online political advertising to combat misinformation. This contributes to the broader scientific understanding of how social media and online advertising can be used to manipulate public opinion and undermine democratic processes.

As the 2024 presidential election campaign looms, along with a record-breaking number of other elections around the world also taking place this year, the work of the Institute for Democracy, Journalism, and Citizenship at Syracuse University takes on a new level of urgency and importance. With social media playing an ever-increasing role in shaping public opinion and political discourse, the need for tools and methods that can help us make sense of this complex landscape has never been greater.

The IDJC’s research, powered by the revolutionary potential of graph technology and exemplified by Neo4j, represents a significant step forward in our ability to uncover hidden relationships and patterns in social media data’s vast and intricate world. By providing a way to map and analyze the connections between ads, funders, candidates, and other key players in the political advertising ecosystem, this work has the potential to shed new light on the forces that shape our democracy.

For those interested in learning more about the power and potential of graph technology, the IDJC’s research is an inspiring example of how this innovative approach can be applied to solve complex, real-world problems.

Learn more about Neo4j and the power of graph technology here.

This blog was written by the Neo4j Query Languages Standards and Research Team¹.

A new standard for a property graph database language, ISO/IEC 39075 Information technology — Database languages — GQL, has been published².

This new standard was developed by the international standards committee, SC32 WG3³, which is also responsible for developing and enhancing the SQL database language standard.

The Neo4j LANGSTAR (languages, standards, and research) team has been actively participating in the development of the GQL standard since the project began.

This post is a short summary of our (as in LANGSTAR) involvement in what is quite a unique experience. New query languages don’t come very often. The last database query language standard developed by ISO was SQL!

What’s in GQL?

In a nutshell, GQL fuses ideas from industry-proven graph query languages, like openCypher, GSQL, and PGQL, with SQL, the foundational language of the database industry, into a full new database language standard, based on eight key ideas:

1. Querying, updating, and managing graph databases using the property graph model.
2. User-friendly and familiar language syntax.
3. Consistent use of visual ‘ASCII-art’ style graph patterns that have been so successful in openCypher.
4. Natural composition of complex queries, such as read-write-read queries (via reading direction-aligned linear statement flow in the style of “MATCH … RETURN …”).
5. Flexible management of multiple graph data products in a central data catalog.
6. Gradual schema design enabled by supporting both schema-free and fixed-schema graphs.
7. SQL-compatible data types and expressions extended with support for native nested data and aligned with established technical standards (Unicode, IEEE 754, ISO 8601).
8. A language foundation that is useful on its own, ready for incorporating additional forms of data into the property graph paradigm, and upon which future versions of the standard can be built.

We will get more into the details on GQL in future blog posts.

How Is an ISO Standard Created?

International standards are created by people who have interest and expertise in the topic being standardized. In the case of SC32 WG3, the participants (individual experts) are delegated by the standards organizations of various ISO member countries around the world. To get involved in the international committee, one must first join a standards organization in some country. The process for joining varies depending on your country.

Working on a standard takes a fair amount of time and effort so, in practice, most individual participants work for companies whose expertise and interest is in data and databases. Experts set their commercial differences aside (but not their opinions!) for the advancement of the space. Standards participants also have to have a tolerance for acronyms and some amount of bureaucracy.

Over more than 40 years of creating database language standards, SC32 WG3 has developed a certain style and culture. Database language standards carefully specify (sometimes in exhausting detail) the syntax and semantics of the language. WG3 builds consensus on the content of the standards using detailed written papers and discussions during meetings. When WG3 participants accept the paper, the GQL editors integrate the changes into the next version of the GQL draft. If a paper is rejected, the author(s) may revise it and bring it back at a later date or may abandon the ideas completely. In any case, we have worked to create an environment where we can argue about the technical ideas during the meeting, and have dinner together in the evenings.

The official project to produce the GQL standard was initiated in 2019 with a New Work Item Proposal (NWIP). The NWIP was initiated by WG3 in June 2019 and approved by the national body participants in September 2019. We had done some amount of work preparing for the GQL NWIP, so it took about five years to complete the GQL standard.

It Takes a Lot of Meetings to Make a Standard!

Prior to 2020, WG3 held two or three in-person meetings a year in various locations around the world. In 2020, we met in person once in January, before face-to-face meetings went out of style, for reasons any reader can probably relate to. When it became clear that we were not going to be able to have face-to-face meetings, we adapted and moved to shorter but more frequent web conferences. Since WG3 is an international standards committee, we have participants from Japan, Korea, China, sometimes Australia, mainland Europe, the UK, and the US. With this distribution of participants, there is no single time that works for everyone. So, we rotated the start time for the web conferences so that everyone had a lousy time of day at some point. The 38 meetings to produce the GQL standard included eleven face-to-face meetings and twenty-seven web conferences.

In the US, there were also two-hour expert group meetings every other Tuesday from 2019 through 2023 plus lots of work and async communication in-between sessions.

GQL Trivia

How big is GQL?

628 pages in total. This is about the same number of pages as SQL-92, which was not the first, but the second major revision of SQL.

How many papers does it take to make a standard?

In addition to the main spec, the GQL standard incorporates 430 papers that were developed, reviewed, discussed, and accepted into the GQL standard.

430 papers across 38 meetings works out to a dozen or so GQL papers per meeting. However, while the GQL standard was undergoing development, we were also developing the 2023 edition of the SQL standard. A key addition here was in a part of the new SQL standard that overlaps with GQL. Among other things, this defines rules for mapping tables to graphs, and adds GQL syntax for matching patterns. (In case we didn’t mention it, the same Neo4j team that focused on the GQL standard also contributed to adding graph pattern matching to the SQL/PGQ standard.)

Fun⁴ statistics

The most papers at a single meeting was in February 2023 where we reviewed and accepted 85 GQL papers during a five-day meeting.

The length of a GQL paper varies. WG3 change proposals include some introduction, descriptive text, examples, the proposed change to the current draft, and some additional stuff such as references and a checklist. It is difficult to do anything in less than about three pages. Ten to twenty pages is common, a number of papers exceeded 50 pages, and a much smaller number exceeded 100 pages. A paper on graph types stands out. The paper was a total of 177 pages, although the last 100 pages were examples illustrating the results of the proposed changes.

The longest of the 430 GQL papers: 177-page paper on Graph Types

Who Did the Work?

Creating a standard is a collaborative effort – the GQL standard is the result of work by all of the participants.

Organizations that participated in the U.S. GQL Expert Group include Datastax, Google, IBM, Intel, Katana Graph, Neo4j, Optum Technology, Oracle, PuppyQuery, RelationalAI, and TigerGraph. Many of the GQL expert group participants also participated in the international group, WG3. In addition, WG3 had participants from Actian, Boray Data, Cannan Consultancy, CnTechSystems, EDB, Profium, LDBC, TF Informatik, Tokyo Metropolitan University, and University of Edinburgh.

Reflections

From a Neo4j LANGSTAR point of view, working on a database language standard was a steep learning curve. Writing WG3 change proposals is a skill one learns by experience, by reviewing papers written by others, and by being reviewed. The WG3 process requires writing, presenting, reviewing, modifying, and re-presenting. We learned a lot doing that, and in spite (or because) of the long hours and complex technical discussions, it was gratifying fun!

Our colleagues in INCITS GQL Expert Group and WG3 are mostly employed by database vendors, so we’ve gained extensive experience playing well with others (often in odd hours, due to coordinating meetings between experts from around the globe). In a world where it looks increasingly impossible to agree on anything, we have shown that even business competitors can work together to make something good for their customers. That was refreshing and very rewarding.

Additional Information

To learn more, the following blogs and documents provide additional information about the GQL standard, Neo4j Cypher, and openCypher:

• ISO GQL: A Defining Moment in the History of Database Innovation
• ISO/IEC JTC 1 GQL Database Language
• GQL: The ISO Standard for Graphs Has Arrived
• GQL is Here: Your Cypher Queries in a GQL World

¹ The Neo4j Query Languages Standards and Research Team includes Finbar Good, Keith Hare, Stefan Plantikow, and Hannes Voigt.

² As of April 11, 2024.

³ The full designation is “ISO/IEC JTC1 SC32 WG3 Database Languages” where ISO is the International Organization for Standardization and IEC is the International Electrotechnical Commission. JTC1 is a Joint Technical Committee underneath ISO and IEC. JTC1 is responsible for most of the computer related standards. SC32 is SubCommittee 32, which is responsible for data management and interchange standards. WG3 is Working Group 3, which is responsible for database language standards, namely the SQL standard and now the GQL standard.

⁴ Standards people might have a weird sense of fun.

After several years of work, the GQL standard has been published. The path started by ISO in September 2019 hit its last major milestone in March 2024, with unanimous approval of the Final Draft International Standard (FDIS) ballot for GQL. This is a very exciting moment for everyone involved with graph databases. SQL is no longer the only ISO standard for database query languages: it now has a younger (and better-looking) sibling.

Two initial questions will likely spring to mind for any graph database practitioner.

The first question is easy to answer: no, GQL has nothing to do with GraphQL, in the same way that GraphQL has nothing to do with graphs. The unfortunate/cheeky/clever (pick one) name clash is just the first hurdle to overcome.

The second question requires a longer answer. Neo4j is fully committed to making the GQL standard a resounding success for the graph database industry. To do that, we want to make the transition as painless as possible for our customers. What will happen to my Cypher queries? is an important question to answer and is the subject of this blog post.

How GQL Impacts Cypher

Cypher is the property graph query language created by Neo4j. The intent at the time was to emulate SQL where possible but innovate where necessary. Its specification was open-sourced via the openCypher project around 2015 and implemented by several other graph products. It is undoubtedly the current de facto standard for property graph query languages. The overwhelming majority of graph database users write queries in Cypher. This blog post focuses on Neo4j’s proprietary Cypher implementation. A future post will discuss the future of openCypher project in the age of GQL. The GQL ISO standard drew inspiration from several existing languages (see the GQL Manifesto), with Cypher/openCypher being significant influences. As a result, Cypher and GQL are quite similar. Over the years, Cypher has proven its power as a graph query language through real-world usage. GQL has built upon Cypher’s strengths, incorporating tweaks to better align with SQL and ensure its long-term viability as a database language. Some of these improvements are exciting, and we’ll discuss them further below.

To provide users with the smoothest possible transition, Neo4j has decided to organically evolve the Cypher language toward GQL compliance: Neo4j will not offer a separate alternative query language to Cypher, but will make Cypher a GQL-compliant implementation.

The GQL standard, like the SQL standard, does not prevent language extensions. Anything not covered by the standard is fair game for implementers, much like it is for SQL. Keep in mind that this is the first version of the GQL standard, which, from a standing start, had to cover a lot of ground. Not every desired language feature made the cut, and despite that, the GQL standard is more than 600 pages, supported by over 430 technical papers (just to put the numbers in context, the GQL:2024 standard is approximately the same number of pages as SQL-92). Big standards are hard to implement, and the wide availability of good implementations is an essential measure of success.

There are features in Cypher that did not make it into the standard and might come up in a future standard release, or not. Still, those Cypher features will remain available to Neo4j users and continue to be fully supported as part of our overall commitment to supporting Cypher. Their use has no negative impact on the GQL compliance of Neo4j. More Cypher extensions will likely be developed over time.

The GQL standard includes both mandatory and optional features. For an implementation to be considered conformant, it must support all mandatory features. However, the long-term expectation is that most GQL implementations will support not only the mandatory features but also most of the optional ones.

If this or future versions of the GQL standard specify features that aren’t implemented in Neo4j Cypher, we will consider adding those features based on customer priorities, just as we’ve always done with our products.

In summary, Cypher GQL compliance will not stop any existing Cypher query from working and will allow Cypher to keep evolving to satisfy users’ demands.

You Say Potayto, I Say Potahto¹

GQL and Cypher are like different pronunciations of the same language.

GQL shares with Cypher the query execution model based on linear composition. It also shares the pattern-matching syntax that is at the heart of Cypher, as well as many of the Cypher keywords (which as pointed out earlier were originally sourced from SQL). Variable bindings are passed between statements to enable the chaining of multiple data fetching and updating operations. And since most of the statements are the same, many Cypher queries are also GQL queries. As a simple example, the well-trodden Cypher below is also GQL:

MATCH (a:Actor)-[:ACTED_IN]->(m:Movie)
WHERE a.name = 'Tom Hanks'
RETURN m.title 

While working on the ISO standard draft, we started implementing some of the early and exciting GQL features not yet in Cypher. Some of the features were on our to-do list, and the standard gave us the opportunity to implement them in a GQL-compliant way. For example, while we did not call out GQL explicitly at the time of their release, the recent improvements to graph pattern matching like the ability to have WHERE clauses inside node expressions (Neo4j 4.4) or inside relationship expressions (Neo4j 5.0), richer label expressions (Neo4j 5.0), and more sophisticated repetitions of patterns with quantified path patterns (Neo4j 5.9) are all examples of GQL features that have already been added to Neo4j!

We have also implemented some more minor but valuable GQL additions, such as the Unicode normalize() function and normalization predicates (Neo4j 5.17).

We have also started accommodating the GQL standard by offering SQL-like synonyms of the Cypher terminology. For example, you can now (Neo4j 5.18) INSERT nodes and relationships and get the same results you would get if you CREATE nodes and relationships. You will notice that GQL tries to unify some of the terminology with SQL. But don’t despair, the existing Cypher terminology is here to stay, so you can change it if you want to.

And obviously, there are more GQL features on their way.

Some changes will involve Cypher users a bit more. A few GQL features might modify aspects of existing queries’ behavior (e.g., different error codes). As such, we classify these GQL features as possible breaking changes. We are working hard to introduce these GQL changes in the Neo4j product in the least disruptive way possible.

The Future is Bright for GQL and Cypher

In conclusion, Neo4j is committed to making the Cypher implementation GQL compliant and doing that incrementally and non-disruptively. Existing Neo4j users already have access to the majority of the GQL features, and more of them are being continuously added to the Cypher language. Cypher queries will keep working, the language will get better, more powerful, and more GQL-compliant.

GQL is born; long live Cypher!

To learn more, the following blogs and documents provide additional information about the GQL standard, Neo4j Cypher, and openCypher:

• ISO GQL: A Defining Moment in the History of Database Innovation
• ISO/IEC JTC 1 GQL Database Language
• GQL: The ISO Standard for Graphs Has Arrived
• openCypher Will Pave the Road to GQL for Cypher Implementers

A joint letter to graph customers, the graph curious, and the Cypher community:

Last week, the database world reached a significant milestone. The International Organization for Standardization (ISO) published GQL, a new database language standard designed for property graphs. GQL, which stands for Graph Query Language, is the first new ISO database language since the introduction of SQL in 1987. This milestone has been eagerly anticipated by the graph community for many years, with many companies, including Neo4j and Amazon, actively advocating and contributing to its development.

A GQL Standard Makes it Straightforward to Use Graphs

We’ve often quipped that there are far more graph problems than customers who realize that their problem is best handled as a graph. As we enter into the world of generative artificial intelligence (AI), there is an even greater explosion of applications where graphs are critical for getting accurate, reliable, and explainable results quickly, like GraphRAG. With GQL as a standard, practitioners and buyers alike will be able to use graph technologies with even greater confidence.

Cypher is the Best and Fastest Path to GQL

The question you are probably asking is: What does this mean for my skills and code, and what does it mean for Cypher? We’re here with some good news: all of you who use Cypher now have a well-paved onramp to GQL. Because the two languages have been on a natural and deliberate convergence course, your best path to GQL is to simply keep using Cypher as it evolves. You also have our commitment that we will continue supporting Cypher for many years. In other words, you can put away your forklift!

The Core Syntax for GQL and Cypher is Largely Identical

Many aspects of GQL are identical to Cypher. Most critically, the query structure is the same. GQL supports the familiar MATCH … RETURN statements, and uses ASCII art to describe graph patterns. Likewise, GQL uses the same basic expressions, linear composition, and more.

Where Differences Exist, Support for the New GQL Form Will Be Added

Certain things are different in GQL than in Cypher. For example, GQL uses the keyword INSERT to add a node or relationship to the graph, whereas Cypher uses CREATE. Similarly, a new FOR statement in GQL does the equivalent of UNWIND. In cases such as these, the current Cypher language will remain supported, and we will also be adding support for the GQL variation, so that you can shift over to the GQL syntax in your own time.

Cypher Functionality Not Yet in GQL Remains Supported as Vendor Extensions

The v1 GQL standard is substantial. It’s about equivalent in size and scope to the SQL 92 standard. (For reference, the first version of ISO SQL was SQL 87.) Even so, not everything in Cypher could make it into the v1 GQL standard. That’s OK. GQL will get there. In the meantime, GQL allows for vendor extensions. Therefore, a lot of what you’re using today is OK. Over time, we expect that commands currently in our products but not in GQL will be proposed for the GQL standard. Examples of commands in this category are MERGE, FOREACH, and LOAD CSV.

New GQL Functionality Offers Opportunities for New Capabilities

Finally, some great new functionality is available in GQL, such as quantified path patterns, which provide advanced pattern matching. Again, these are great additions for graph querying, and some vendors have already implemented parts of this. This work will continue, and we will add all of this to openCypher over time so that users looking for a straightforward path to GQL can gain access to this great functionality.

As for differences, there are some, but they are few, and the ones that exist aren’t that significant. These will be handled on a vendor-by-vendor basis, with clear compatibility flags, lots of advance notice, and deprecation flags in the way that’s typical for major versions.

Paving the Way with openCypher

Many vendors who implement Cypher are doing so with openCypher. This open source framework provides an existing foundation of tools and tests for implementing Cypher inside of a product, be it a database or a tool. To help smooth and democratize the transition, openCypher will be providing new artifacts to align with GQL. This will then make its way into our product roadmaps in alignment with your needs: our customers and community.

This path is possible for a few reasons. First, it continues a trajectory of convergence work that has been ongoing for years. Not only was Cypher a major input into GQL, but we have been evolving the language concurrently with GQL so that the two can closely align. Another reason this makes sense is that many of the core individuals actively involved in the ISO GQL standard were (and are) also active in openCypher. All of this puts us in a great position to give you a smooth path to convergence.

Getting to GQL

We will do this in alignment with your priorities and needs, working backward from the customer, to make Cypher an implementation of GQL—in our products and in openCypher. You will see the term GQL show up increasingly often where it makes sense. You will retain the flexibility to use both Cypher and GQL syntax styles when you want, and the term Cypher isn’t going to be going away anytime soon. You will get the best of both worlds: a seamless multi-year transition with lots of optionality, and a strong and familiar path towards all the benefits of a formal international standard.

Cypher to GQL: Practical Considerations

By definition, a database language standard transcends any and all details about an implementation. However, as the world continues its rapid transition to the cloud, we would be remiss in not saying some words about this increasingly common deployment scenario. For the large and growing number of you who are using, or who want to use, a managed graph database service, the transition to GQL will be even more seamless. The nature of a managed platform is that new syntaxes and features will automatically show up for you to use, speeding your path to GQL.

Generative AI Impacts

A blog post would not be complete in 2024 without a mention of generative AI. As knowledge graphs increasingly become a part of the generative AI stack, a soft transition lets you get better usage from large language models (LLMs). They have, after all, already been trained on over 10 years of Cypher examples that are spread throughout the internet. As the term GQL becomes increasingly used alongside Cypher, models will gradually evolve to understand both.

Wrapping It All Up

Thanks for being on the journey with us! With a formal ISO GQL standard as an added wind to all of our sails and a clear and smooth transition path for skills, certification, and query code, you will be able to continue to use your investments, while also benefiting from the power of the new standard. We are excited for this big step, which makes the graph community an even more exciting place to be.

Philip Rathle, CTO Neo4j
Brad Bebee, GM AWS, Amazon Neptune

This post is a joint collaboration between Neo4j and AWS and is being cross-published on both the Neo4j blog and the AWS Database Blog.

To learn more, the following blogs and documents provide additional information about the GQL standard, Neo4j Cypher, and openCypher:

• ISO GQL: A Defining Moment in the History of Database Innovation
• ISO/IEC JTC 1 GQL Database Language
• GQL is Here: Your Cypher Queries in a GQL World
• openCypher Will Pave the Road to GQL for Cypher Implementers

Engaging in a conversation with a company’s AI assistant can be frustrating. Chatbots give themselves away by returning generic responses that often don’t answer the question.

This doesn’t have to be the case. Imagine a different scenario: interacting with a chatbot that provides detailed, precise responses. This chatbot sounds like a human with deep institutional knowledge about the company and its products and policies. This chatbot is actually helpful.

The second scenario is possible through a machine-learning approach called Retrieval-Augmented Generation (RAG).

RAG is a technique that enhances Large Language Model (LLM) responses by retrieving source information from external data stores to augment generated responses.

These data stores, including databases, documents, or websites, may contain domain-specific, proprietary data that enable the LLM to locate and summarize specific, contextual information beyond the data the LLM was trained on.

RAG applications are becoming the industry standard for organizations that want smarter generative AI applications. This blog post explores RAG architecture and how RAG works, key benefits of using RAG applications, and some use cases across different industries.

Why RAG Matters

Large language models (LLMs), like OpenAI’s GPT models, excel at general language tasks but have trouble answering specific questions for several reasons:

• LLMs have a broad knowledge base but often lack in-depth industry- or organization-specific context.
• LLMs may generate responses that are incorrect, known as hallucinations.
• LLMs lack explainability, as they can’t verify, trace, or cite sources.
• An LLM’s knowledge is based on static training data that doesn’t update with real-time information.

To address these limitations, businesses turn to LLM-enhancing techniques like fine-tuning and RAG. Fine-tuning further trains your LLM’s underlying data set while RAG applications allow you to connect to other data sources and retrieve only the most relevant information in response to each query. With RAG, you can reduce hallucination, provide explainability, draw upon the most recent data, and expand the range of what your LLM can answer. As you improve the quality and specificity of its response, you also create a better user experience.

How Does RAG Work?

At a high level, the RAG architecture involves three key processes: understanding queries, retrieving information, and generating responses.

Before implementing an RAG application, it’s important to clean up your data to make it easy for the RAG application to quickly search and retrieve relevant information. This process is called data indexing.

Frameworks like LangChain make it easy to build RAG applications by providing a unified interface to connect LLMs to external databases via APIs. Neo4j vector index on the LangChain library helps simplify the indexing process.

1. Understanding User Queries

The process begins when a user asks a question. The query goes through the LLM API to the RAG application, which analyzes it to understand the user’s intent and determine what information to look for.

2. Information Retrieval

The application uses advanced algorithms like vector similarity search to find the most relevant pieces of information in the company’s database. These algorithms match vector embeddings based on semantic similarity to identify the information that can best answer the user’s question.

3. Response Generation

The application combines the retrieved information with the user’s original prompt to create a more detailed and context-rich prompt. It then uses the new prompt to generate a response tailored to the organization’s internal data.

What Are the Benefits of RAG?

Out-of-the-box generative AI models are well-equipped to perform any number of tasks and answer a wide variety of questions because they are trained on public data. The primary benefit of using a RAG application with an LLM is that you can train your AI to use your data—and this data can change based on what’s most relevant and current. It’s just a matter of which data stores are accessed and how often the data within them is refreshed. RAG also allows you to access and use custom data without making it public. Overall, RAG allows you to provide a generative AI experience that is personalized to your industry and individual business while solving for the limitations of a standalone LLM:

• Increased Accuracy: RAG applications provide domain-specific knowledge and enhanced reasoning, significantly reducing the risk of hallucinations.
• Contextual Understanding: RAG applications provide contextual responses based on proprietary, internal data across your organization, from customer info to product details to sales history.
• Explainability: By grounding responses in a source of truth, RAG applications can trace and cite information sources, increasing transparency and user trust.
• Up-to-date Information: As long as you keep your Graph Database or other document stores updated, RAG applications access the latest data in real time, allowing for continuous improvement.

What Are Common RAG Use Cases?

RAG enhances GenAI applications to interpret context, provide accurate information, and adapt to user needs. This enables a wide range of use cases:

• Customer Support Chatbots: With product catalogs, company data, and customer information at its fingertips, RAG chatbots can provide helpful, personalized answers to customer questions. They can resolve issues, complete tasks, gather feedback, and improve customer satisfaction.
• Business Intelligence and Analysis: RAG applications can provide businesses with insights, reports, and actionable recommendations by incorporating the latest market data, trends, and news. This can inform strategic decision-making and help you stay ahead of the competition.
• Healthcare Assistance: Healthcare professionals can use RAG to make informed decisions using relevant patient data, medical literature, and clinical guidelines. For instance, when a physician considers a treatment plan, the app can surface potential drug interactions based on the patient’s current medications and suggest alternative therapies based on the latest research. RAG can also summarize the patient’s relevant medical history to help guide decisions.
• Legal Research: RAG applications can quickly retrieve relevant case law, statutes, and regulations from legal databases and summarize key points or answer specific legal questions, saving time while ensuring accuracy.

As enterprises continue to generate ever-increasing amounts of data, RAG puts the data to work to deliver well-informed responses.

If you’re assessing your tech stack for generative AI models, the 2024 analyst report from Enterprise Strategy Group, Selecting a Database for Generative AI in the Enterprise, is a valuable resource. Learn what to look for in a database for enterprise-ready RAG applications and why combining knowledge graph and vector search is key to achieving enterprise-grade performance.

NODES, the premier developer conference dedicated to graph-powered applications and contextual AI, returns for its sixth year on November 7, 2024. Join thousands of developers and data scientists in this free, 24-hour conference to learn about the latest GenAI innovations and gain insights from speakers showcasing their implementations, tools, and models.

Last year, NODES 2023 welcomed thousands of graph enthusiasts who attended more than 100 live sessions by 110 speakers from 30 countries. NODES 2024 promises to be even better:

• Sessions will cover hot topics like retrieval augmented generation (RAG) and AI orchestration frameworks
• The conference will span time zones to accommodate our global community
• Two-hour workshops will provide deep dives into graph solutions
• Speakers will present live and interact with attendees during Q&A sessions

Submit Your Talk by June 15, 2024

The call for papers is open now through June 15, 2024. Submit an educational presentation demonstrating graph and Neo4j-related technologies, focusing on solution creation using code, data models, Cypher, or best practices.

Choose from three session formats:

• 30-minute talks with Q&A
• 10-minute lightning talks
• 2-hour hands-on workshops

We accept submissions covering topics in four tracks:

Applications: Libraries, Frameworks, and Platforms

Discover how developers use Neo4j to power inventive solutions across software stacks, cloud providers, and programming languages.

AI: Generative AI, Knowledge Graphs, and Retrieval-Augmented Generation

Explore the intersection of groundbreaking research and real-world applications using graph technologies and techniques.

Data Science: Machine Learning, Graph Data Science, and AI Models

Learn advanced techniques in data curation and maintenance designed to fuel AI models.

Graphs: Visualization, Data Integrations, and Tips & Tricks

Unlock the power of graphs, connect knowledge graphs to broader data systems, and uncover expert tips and tricks.

Get Inspiration From NODES 2023

If you missed NODES 2023 or want to revisit some of the best sessions, check out our blog post featuring the top 10 sessions from NODES 2023. These highly-rated and well-attended presentations cover various topics, from creating graph dashboards with natural language queries to applying graph data and AI for pet travel. You’ll also find sessions on Cypher types, user-defined procedures, GitHub Actions workflows, and more. The full NODES 2023 playlist is available on YouTube here.

Submit your talk by June 15 and save the date for NODES 2024!

Last Friday, ISO published a new database query language: ISO GQL. It’s a peer language to SQL and the first new ISO database language since 1987 — when the first version of SQL was released.

This is a really big deal.

The GQL standard is for and about graphs. Graphs are a way of working with data that you already have. They excel with exactly the type of data that the world is creating more of, building applications around, and increasingly finding ourselves needing to analyze and make sense of.

Example of a graph visualization

Database management systems (DBMSs) that work with your data as a graph are called graph databases. The ISO GQL (Graph Query Language) standard uses a particular type of graph called a property graph. Its design dates back over two decades, and it has gained significant popularity in recent years across a wide range of applications – spanning a multitude of use cases in just about every industry.

Having an international standard for graph databases adds immense value to a landscape where data is increasingly dynamic and interconnected. That ISO has invested more than five years in creating this standard says something about the importance of this technology.

Why Graphs, Why Now?

Standards channel trends. Graph databases are increasingly used by a wide variety of organizations to solve all kinds of problems. Just this year, Gartner placed knowledge graphs, something for which graph databases are purpose-built, at the center of their 2024 Technology Impact Radar (which ranks the 30 most impactful technologies) and their 2024 Impact Radar for GenAI. Neo4j, the leading but very much not the only player in this rapidly expanding space, has amassed the majority of the Global 2000s, countless startups, and major governments as users and customers.

Graphs solve real-world high-value problems that otherwise defy solving, both the operational database world and the world of analytics and AI.

Power of Standards

Standards shape industries. They are where innovation crystallizes into common methods and patterns that can be applied to solve large-scale problems in the real world.

Standards are a friend to the CIO, the developer, and the ecosystem alike. For CIOs, they are the best antidote to vendor lock-in and guarantee access to a large pool of common skills. For developers, they avoid needing to learn new technologies while shuttling between projects and products. And for the ecosystem, they provide clear integration patterns that amplify the value and reach of technology investments. All three are protection against obsolescence.

What does this mean for graphs? One obvious takeaway is that graph databases have now earned their place in the information technology mainstream alongside traditional databases. This is a good cause for re-evaluating what graph databases can do for one’s organization today.

What’s probably not as obvious for many readers is how perfect the timing is. GenAI is a crashing wave that is further accelerating digital transformation. The same things that make graph databases peerless at modeling connected real- and digital-world systems (such as smart cities, payment networks, supply chains, biological systems, asset ownership chains, computer networks, etc.) also make them invaluable for GenAI.

In a world where LLMs provide a cornucopia of value and surprises in what amounts to right-brain behavior for AI, AI engineers at the leading edge of their field are discovering that graph databases can play the essential role of the left brain. This is not a small thing in a world where most problems can almost certainly benefit from having the power of both hemispheres.

Much has been written about this last topic, and I look forward to covering it in greater detail in a separate post.

What Does GQL Look Like?

As a peer, complementary language to SQL that originates from the same organization and committee as the one behind SQL, it won’t surprise you to learn that GQL resembles SQL in many ways. Both languages share the same data types, and many of the same keywords and commands are the same.

Of course, GQL also includes parts that cater specifically to the unique aspects of graph databases. The absolute core of the language is the ASCII-art-inspired way of defining patterns. A few graph database languages share this, and it traces its way back to 2010 with Neo4j’s Cypher language, which, through the openCypher project, has become the de facto standard graph query language used by Neo4j, AWS Neptune, and numerous others.

Path to GQL

In the world of graphs, there can be many ways to reach a destination. Often, the shortest path can be made clear. Such is the case with GQL. There is much more to be said about this – and it will be the subject of future posts. But the short answer is that the shortest path to GQL is Cypher, which most people already use.

There are several reasons for this:

First, Cypher was a major input into GQL. And as inputs go, it was of especially high quality. By the time the standards work kicked off in 2019, Cypher had already undergone nearly a decade of real-world trial-by-fire maturation.

Second, Cypher itself was originally modeled after SQL. The “no idle variance from SQL” principle that drove Cypher’s development from the start turned out to be prescient. In the early days of Cypher, no one would have imagined that it would become a significant input to an ISO standard – let alone on a convergence course. But the principle made sense. Why invent something new to do something people are already used to doing?

There is one more reason: the team behind Cypher and openCypher has been deeply involved in developing the GQL standard. Around half a dozen Neo4j engineers collaborated with numerous others throughout the roughly five years it took to produce GQL, joining up as full-time standards committee members. As aspects of the GQL standard crystallized, Cypher itself was made to align with the forthcoming standard. The roads were smoothed in anticipation of GQL.

As a result, today’s Cypher is already highly aligned with GQL. New and existing graph database users alike can benefit from a clear path to adoption and a smooth path to GQL. Sticking with Cypher will land you straight into GQL as it completes the last few steps of its convergence course. So, if you got worried and brought out your forklift when you first saw this post, you can now put it away!

Closing Thoughts

This is a massive milestone for the database industry, one whose impact will be felt over years and decades, not just months. Standards, after all, are meant as a foundation for enduring technologies and are reserved for only the most important trends.

We’re excited about the possibilities and look forward to seeing what innovation and disruption you will all create in your respective spaces thanks to graphs – now with GQL at your backs!

To learn more, the following blogs and documents provide additional information about the GQL standard, Neo4j Cypher, and openCypher:

• ISO/IEC JTC 1 GQL Database Language
• GQL: The ISO Standard for Graphs Has Arrived
• GQL is Here: Your Cypher Queries in a GQL World
• openCypher Will Pave the Road to GQL for Cypher Implementers

GETTING STARTED WITH GRAPHS

• GRAPHACADEMY: Introduction to Vector Indexes
• READ: Graph Search Algorithm Basics
• WATCH: Introduction to Neo4j
• TRY: Neo4j AuraDB Free

• FEATURED COMMUNITY MEMBER: Astrid Krickl

  Astrid is working at Semantic Web Company as a Data and Knowledge Engineer. At WU Vienna, she investigated tools and techniques (mainly NLP and machine learning) that can help with the problem of misinformation and fake news. Connect with her on LinkedIn. In a livestream “Building a Semantics-Based Recommender System for ESG Documents” we brought together Semantic Web technologies and Neo4j to build a knowledge-based recommender system that is powered by a knowledge graph based on ESG-related documents.

  
   

  PODCAST: Providing Better Business Intelligence with Vish Puttagunta

  For April, our Podcast is joined by Vish Puttagunta, the CEO of Power Central, where they are bringing ERP, data & intelligence together in the food industry to generate tangible and measurable ROI in Business.
   

  RAG: Neo4j & Haystack Part 1: Knowledge Graphs for RAG

  Neo4j and Haystack teamed up for a two-part webinar series to discuss Knowledge Graphs and RAG. In Ep1, Andreas Kollegger covers knowledge graphs and how to use them in LLM applications and RAG. And in Ep2, Sergey Bondarenco shows you how to use the Haystack Neo4j integration to make full use of knowledge graphs in Haystack pipelines.
   

  BOOK: Neo4j: The Definitive Guide

  This book by Luanne Misquitta and Christophe Willemsen is now available as an Early Release. It is a comprehensive guide to mastering Neo4j enterprise deployments. Drawing from real-world experiences, it offers practical advice on improving Cypher queries, data modelling, security, and observability.
   

  ERP: Navigating the Complexities of PLM-ERP Integration with Graph Models and Modern Digital Product Data

  The article by Oleg Shilovitsky discusses the complexities of integrating PLM and ERP systems in manufacturing, highlighting graph-based models for better data management and integration. It emphasises the need for new approaches to manage intricate data and process interdependencies, suggesting that modern graph models can significantly ease enterprises’ integration challenges.

  TWEET OF THE WEEK: Hanley

  Sorry, knowledge graph! Neo4j in particular. Their DB is absolutely amazing. It has it all.

  — Hanley — / (@PaperWizardAI) March 19, 2024

  Don’t forget to share it if you like it!
   

We’re thrilled to announce new native integrations with Google Cloud and Vertex AI that solve a critical challenge in GenAI development: accessing contextually rich external data to deliver accurate, explainable results. The integrations streamline the implementation of GraphRAG, a responsive and real-time method to address these issues.

GraphRAG combines two powerful technologies: retrieval-augmented generation (RAG) and knowledge graphs. RAG allows GenAI applications to access and query external datasets, while knowledge graphs make the data smarter by enriching the contextual information with entities and capturing the complex relationships between them. This enriched context enables LLMs to reason, infer, and accurately answer questions and execute tasks, anchoring their responses and actions in factual information.

The importance of knowledge graphs in GenAI development cannot be overstated. Gartner considers knowledge graphs essential to the development of GenAI and has urged data leaders to “leverage the power of LLMs with the robustness of knowledge graphs to build fault-tolerant AI applications.”

Available now, the GraphRAG integrations allow organizations to rapidly build GenAI applications that can integrate contextually rich external data in real time while being secure and compliant. This capability dramatically reduces hallucinations while enabling LLMs to uncover and use complex relationships and patterns within large datasets—so GenAI apps can deliver the accuracy, relevance, and explainability required by enterprise use cases.

The integrations enable developers to implement GraphRAG seamlessly:

1. Quickly create knowledge graphs for accurate, explainable results. Developers can easily create knowledge graphs using Gemini models, the Google Cloud VertexAI platform, LangChain, and Neo4j from unstructured data like PDFs, webpages, and documents—either directly or loaded from Google Cloud Storage buckets. The simplified process uses llm-graph-transformer, which Neo4j contributed to LangChain. It provides LLMs with the intended graph schema and uses Gemini’s function-calling capabilities to extract entities and their relationships in a structured way. These entities and relationships are then added to the Neo4j knowledge graphs for use in GenAI or other applications. With GraphRAG, LLMs can access data modeled around entities, their attributes, and the relationships between entities, improving GenAI accuracy and explainability.


2. Ingest, process, and analyze real-time data in seconds. Developers can use Flex templates in Dataflow to create repeatable, secure data pipelines that ingest, process, and analyze data across Google BigQuery, Google Cloud Storage, and Neo4j—supplying knowledge graphs with real-time information and enabling GenAI applications to provide relevant, timely insights.


3. Build and deploy graph-powered GenAI apps with Gemini for Workspace and Reasoning Engine. Deploying GenAI applications to production has been challenging, but with Reasoning Engine from the Vertex AI platform, developers now have the tools to easily deploy, monitor, and scale GenAI apps and APIs onto Google Cloud Run. Neo4j’s GenAI capabilities, such as Vector Search, GraphRAG, and conversational memory, integrate seamlessly through LangChain and Neo4j AuraDB on Google Cloud. The Gemini models have been trained on Neo4j-specific assistant content, streamlining development by automatically generating code snippets for Neo4j tools and APIs. It also translates natural language into Neo4j’s Cypher query language, making it easier to build applications with Neo4j. With just a few lines of Python code and the Google Vertex AI Python SDK, you can deploy GenAI and graph-powered APIs to Reasoning Engine, making powerful RAG capabilities available for both developers and operations teams. This empowers organizations to bring graph-powered GenAI applications to production with ease and confidence.

Let’s dig a little more deeply into the new integrations and how they help organizations realize the immense potential of GenAI.

Constructing Knowledge Graphs From Unstructured Data With Gemini Models and LangChain

Gemini’s advanced language capabilities and new function-calling features enable it to identify entities, types, and relationships from unstructured text and extract them in a structured way. Using the llm-graph-transformer that Neo4j contributed to LangChain, developers can turn any set of LangChain documents— PDFs, web pages, Google Docs, and more—into a knowledge graph.

By providing a specific prompt with the instruction and an optional graph schema with the unstructured text, developers can guide the LLM to extract information and populate structured output via pre-defined objects for nodes and relationships. With Gemini’s initial support for function calls, developers can use the model in the extraction pipeline.

Powering Real-Time GraphRAG Applications With Dataflow Flex Templates

Our new integrations allow developers to build GraphRAG applications using real-time data. They can use Dataflow Flex templates to set up secure, efficient pipelines for real-time data movement from BigQuery and Google Cloud Storage into their Neo4j Graph Database on Google Cloud. The knowledge graphs powering GraphRAG applications are continuously updated, enabling them to generate more accurate, timely, and contextually relevant responses.

Architecture diagram showing Dataflow from Google Cloud to Neo4j

Below, we use Dataflow Flex templates to run a job from the Google Cloud Storage bucket and transform the data into a Neo4j knowledge graph. As the data is updated in Google Cloud, we can create a job that checks for changes and updates the knowledge graph in real time. Now the graph within Neo4j can be used with RAG architecture for greater LLM accuracy and explainability.

A sample job run in Dataflow to get data from Google Cloud to Neo4j to create a knowledge graph

Organizations can combine Dataflow Flex templates with Neo4j’s graph database to build sophisticated GraphRAG applications that adapt to rapidly changing data landscapes. With real-time data, these applications deliver more accurate, timely, and contextually rich insights, enhancing decision-making and user experience across domains and use cases.

Accelerating GenAI With Gemini for Google Workspace and Neo4j

We’ve worked closely with Google to improve the graph application development capabilities of Gemini for Google Workspace. Neo4j has provided extensive training data to Google, enabling Gemini to understand the Cypher query language and provide more comprehensive guidance to graph application developers. (The training data included Neo4j’s documentation, online courses, and knowledge base, as well as data from our text2cypher development efforts and crowdsourced, LLM-generated question-answer pairs.)

The Gemini for Google Workspace developer assistant can now help developers create knowledge graphs in Neo4j by translating natural language into Cypher. Applications can be integrated for vector search capabilities with knowledge graphs and augmented with GraphRAG to ground LLMs for more explainable and accurate results. Once knowledge graphs are created within Neo4j, they can be used to explore graph data to uncover hidden patterns and insights.

A code snippet within a Google Workspace, showcasing the generation of Cypher query to load data into Neo4j

Gemini for Google Workspace is also trained in integrating Neo4j with popular orchestration frameworks like LangChain, LlamaIndex, and Haystack, so it can offer framework-specific guidance to developers, further streamlining the development process.

Gemini for Google Workspace is available for end users in Google Workspace and developers in the Google Cloud Platform (GCP) console, as well as in popular development environments like Visual Studio Code and JetBrains. This allows developers to tap into AI-assisted coding capabilities as they build the next generation of GenAI-enabled applications.

Deploying Graph-Powered GenAI Applications With Google’s Reasoning Engine Runtime

Many developers are new to deploying GenAI applications in production environments. Google’s Reasoning Engine Runtime addresses this challenge by simplifying the process of securely deploying, scaling, monitoring, and operating GenAI applications with Vertex AI and Gemini models. Google’s Reasoning Engine Runtime is a new product that goes beyond the capabilities of Vertex ML by providing a framework for integrating knowledge graphs with GenAI applications. It helps developers decide when to use a knowledge graph based on the specific requirements of their application, such as the need to model complex relationships between entities or perform advanced reasoning tasks.

Our new integrations with Google Cloud, combined with our extensive LangChain integrations, allow users to seamlessly incorporate Neo4j knowledge graphs into their GenAI stack. Developers can use LangChain to run direct or advanced RAG architectures, including GraphRAG, within Reasoning Engine Runtime.

Combining Neo4j’s knowledge graph capabilities with Google’s Reasoning Engine Runtime is a powerful approach to building contextually advanced GenAI applications. It reduces the complexities of productionization while delivering more accurate and explainable GenAI results.

GraphRAG: Unlocking the Potential of GenAI

As GenAI has evolved, it’s become increasingly clear that GraphRAG is a powerful tool for overcoming the limitations of LLMs. Combining knowledge graphs with retrieval-augmented generation resolves the critical issues of accuracy, explainability, and transparency—and unlocks the full potential of GenAI.

The new integrations between Neo4j and Google Cloud make GraphRAG more accessible and simpler to use than ever before. Now, instead of struggling to address hallucinations or lack of transparency, developers can focus on creating a new generation of reliable, context-aware GenAI applications.

SOC2 Type 2 Brings 3 Major Benefits to Enterprise Organizations

1. There are many benefits that SOC2 Type 2 compliance brings to enterprise organizations. First and foremost, it assures that their data is being managed in a secure and compliant manner. This is particularly important for companies that handle sensitive or regulated data, such as healthcare or financial data.


2. SOC2 Type 2 compliance demonstrates that Neo4j has implemented robust security controls and processes to protect customers’ data. This includes measures such as access controls, data encryption, and regular security audits and testing.


3. SOC2 Type 2 compliance is an important factor in maintaining customer trust and confidence. With data breaches and cyber-attacks becoming increasingly common, enterprise organizations must work with service providers who take security and compliance seriously. In achieving SOC2 Type 2 compliance, Neo4j is demonstrating its commitment to protecting our customers’ data and ensuring the highest levels of security and compliance.

GETTING STARTED WITH GRAPHS

• GRAPHACADEMY: CSV Import
• READ: Why Graph Technology is the Future
• WATCH: Introduction to Neo4j
• TRY: Neo4j AuraDB Free

• FEATURED COMMUNITY MEMBER: Saurav Joshi

  Saurav Joshi is a passionate data scientist with expertise in data mining, statistics, NLP, causal inference, reinforcement learning, optimisation, LLMs, and synthesising AI solutions with cloud platforms to drive impactful solutions. Connect with him on LinkedIn. In a livestream “Enhanced QA Integrating Unstructured Knowledge Graph Using Neo4j and LangChain” we talked through a project that uses the robust capabilities of Neo4j Vector Index and Neo4j Graph Database to implement a retrieval-augmented generation system, aiming to provide precise and contextually rich answers to user queries.

  
   

  GENAI: GenAI Ecosystem

  The Neo4j GenAI Ecosystem is a collection of tools and integrations that make it easy to use LLMs with Neo4j. It is your one-stop resource for Graph RAG, GenAI Knowledge Graphs and more.
   

  WORKSHOP: Videos now available on-demand

  Last week, we completed our set of workshops and this time, we covered Introduction to Neo4j, Liquibase with Neo4j, GenAI Deployment Best Practices and Geospatial Data Analytics. All recordings are now available for you to watch.
   

  BOOK: Graph Data Science with Python and Neo4j

  Timothy Eastridge’s new book is a comprehensive guide that teaches you to enhance data analysis and insights by integrating Python, Neo4j and advanced technologies, including LLMs.
   

  GRAPH DATABASE: My passion for the world of graph databases

  Michal Štefaňák develops the PHP Bolt Driver as well as cypherGUI. He thrives on making graph databases more approachable with his tools and his active involvement in the community. In this blog, he shares his passion for Graph Databases.

  TWEET OF THE WEEK: Chris

  Just built a working proof of concept using #DSPy to model text and build a Knowledge Graph using @neo4j . The schema is included as context in the prompt so the LLM knows how to make connections with existing data. Repo in reply. pic.twitter.com/CrmOxRKjSL

  — Chris (@chrisammon3000) March 28, 2024

  Don’t forget to share it if you like it!
   

We’re thrilled to collaborate with Microsoft on a unified data offering to uncover hidden data patterns and address critical GenAI challenges, such as the ability to provide personalization and deliver relevant search outcomes through additional context.

Neo4j’s graph capabilities—knowledge graphs, data science algorithms, graph-powered RAG, and vector and semantic search— will now be natively integrated into the Microsoft Fabric analytics platform and Microsoft Azure OpenAI Service.

Customers using Neo4j, Microsoft Fabric, and Azure OpenAI together can seamlessly combine structured and unstructured data, easily discover hidden patterns across billions of data connections, enhance contextual understanding within their data, and rapidly deliver enterprise-grade GenAI applications.

The new offering allows organizations to:

• Transform unstructured data into knowledge graphs. Azure OpenAI can process unstructured data and load it into a knowledge graph, allowing Neo4j query tools to extract powerful insights.
• Enhance contextual understanding and explainability with GraphRAG. GraphRAG applications, fully integrated with Neo4j’s GenAI functions and Azure OpenAI, can use knowledge graphs derived from enterprise data to enhance query prompts dynamically.
• Provide long-term memory for LLMs with vector embedding integration. Neo4j supports native vector embeddings, and developers can use OpenAI embedding APIs to create embeddings and store them in the Neo4j database.
• Generate graph-powered insights as part of Fabric. Fabric customers can now use Neo4j graph analytics capabilities to quickly find hidden patterns and relationships within their data.
• Deploy Neo4j Graph Analytics as a native Fabric workload. Neo4j Graph Analytics will soon be a native Microsoft Fabric workload, allowing users to create graph models from OneLake data, run Graph Data Science algorithms, and write results back into OneLake.

Here’s a more detailed look at the integrations, starting with Neo4j + Azure OpenAI and moving on to Neo4j + Microsoft Fabric.

Neo4j and Azure OpenAI: Graph Insights and Enterprise-Ready GenAI

Since 2009, Neo4j has pioneered graph databases, helping organizations find meaning in complex, interconnected datasets. Graph technology is set to transform enterprise analytics and AI, with Gartner predicting that it will drive 80% of data and analytics innovations by 2025—up from 10% in 2021. Gartner also sees knowledge graphs as “a vital capability” and “the first step to resolving fragmented data management issues by enabling a GenAI-augmented data fabric.”

Neo4j has been working with Azure OpenAI Service since its private preview. Combining the tools allows us to build graphs from data we might not be able to ingest and model otherwise, and to make those graphs accessible to a much broader group of people.

Any industry with underused, highly connected data can benefit from this approach. Neo4j and Azure OpenAI can support use cases across the financial service industry, for example, fraud and money laundering detection. Supply chain and manufacturing use cases include knowledge transfer, bill of materials management, and applications for optimization.

From our first project with OpenAI onward, we’ve observed a common architecture for applying GenAI with graphs:

Neo4j’s Knowledge Graph and Generative AI reference architecture

This architecture consists of:

1. Ingestion – Extracting a knowledge graph from structured, semi-structured, and even unstructured data using the Azure OpenAI Service, then feeding it into the Neo4j Graph Database. Source data might reside in Fabric, Azure Blob Storage, or elsewhere. Automating ingestion with generative AI reduces the cost of getting started with graph databases, making it possible to gain value from connections in data where previously impossible.
2. Consumption – Before generative AI, it required deep expertise to interact with graphs and get value from the connections in data. Layering Azure OpenAI Service over the Neo4j enables any user to interact with a graph.

Consider a specific example: a medical case sheet data collection. We’re going to parse that data to build a knowledge graph and then layer a chat interface on top with options to run in a Streamlit application:

This example shows transforming unstructured data to a knowledge graph for consumption

Ingestion and Knowledge Graph Extraction

In this example, we use zero-shot with a simple prompt and the gpt-4-32k model (You can find more information on OpenAI models here). That allows us to extract case sheet information for each person into a Neo4j knowledge graph (Check out the notebook on the GitHub repository for more details). Here’s the resulting data model:

Showing extraction of a case sheet information for each person by a Neo4j data model

Let’s consider what we’ve just accomplished. We’ve used generative AI to build and populate a knowledge graph with unstructured medical case data. This project might take weeks to do manually; with Neo4j and the OpenAI model, we’ve done it in minutes. We can apply graphs to entirely new problems. The vast stores of untapped enterprise data—consider information regarding drug interactions, shipping routes, or data breaches—hold the potential to construct new graphs. This allows users to derive value from connections within their data that were previously undiscovered.

Consumption

What’s next now that we have a graph? Neo4j offers many tools to interact with the graph, from the Cypher query language to the Bloom graph visualization tool. Generative AI lets us do something new: interact with the graph by asking natural language questions. One of the simplest ways to do this is with the LangChain integration between both Neo4j and the Azure OpenAI Service:

Showing the flow to convert the natural language prompt to Neo4j Cypher query language

Stepping through this, we see that a user enters a prompt to ask, “How many of my patients suffer from both coughing and weight loss?” That is passed to the chatbot, a Python application that uses LangChain. LangChain passes the prompt to the Azure OpenAI Service, converting the question into a Cypher statement that will address the question. LangChain then passes that Cypher statement to the Neo4j database, and a response is received.

Finally, LangChain invokes the Azure OpenAI Service a final time. The service summarizes the JSON blob resulting from the database query into natural language, which is presented to the user in the Streamlit UI.

So, what have we accomplished? We’ve enabled folks to ask natural language questions to a Neo4j database. That means non-technical users can now explore graphs and get value from the connections in their data—like understanding how many patients suffer from coughing and weight loss. Our customers use this technique to enable internal users to answer questions about their business practices based on a knowledge graph representing internal technical documents.

Neo4j and Microsoft Fabric: A New World of Data Exploration, Analysis, and Collaboration

Microsoft Fabric is a unified platform for data analytics, combining a range of data toolsets under one umbrella—and integrating with Neo4j unlocks its full potential, revolutionizing the data exploration and analysis enabled by Fabric.

In Fabric LakeHouses, data is typically stored as files or rows and columns in SQL Server tables, which may not adequately capture intricate relationships within datasets, impeding thorough analysis and insights.

Neo4j’s Graph Database represents data as nodes and relationships, enabling intuitive visualization and comprehensive exploration of connections within datasets. With built-in machine learning and AI algorithms, Neo4j helps organizations uncover hidden patterns and derive deeper insights from their data, ultimately driving better business outcomes. For example:

• Reduced operational burden on IT teams. IT teams running Neo4j AuraDB within the Fabric ecosystem can uncover hidden data patterns and actionable insights within their data far more quickly and easily.
• More informed business decision-making. By integrating the Neo4j BI Connector with Fabric Power BI, organizations can enable real-time data access across many data streams. Business decision-makers gain a deeper, more comprehensive understanding of data.
• Better customer experience and risk management. Running Neo4j Graph Data Science algorithms with Fabric Data Science allows deeper analysis of technical patterns to improve predictions and create node embeddings, offering insights into individual node relationships—invaluable for customer experience, risk management, and more.

Current and upcoming Neo4j and Microsoft Fabric integrations include:

1. Synapse Data Engineering module integration. By leveraging Python-based notebooks within Microsoft Fabric’s Synapse Data Engineering module, users can tap into Neo4j’s graph data seamlessly. The integration allows data scientists to import Neo4j libraries directly, enabling tasks such as reading, writing, and employing graph data science algorithms effortlessly.
2. Neo4j Browser integration. Neo4j Browser is a developer-friendly interface for executing Cypher queries and visualizing results, facilitating ad-hoc graph queries and prototype development from the browser interface. With support for loading various file formats, including JSON from OneLake, users can easily import and manipulate data, enriching their analysis with Neo4j’s graph insights.
3. Data Factory and Neo4j’s JDBC/ODBC drivers. By using these drivers within Data Factory, organizations can seamlessly transfer data between Neo4j and Fabric, enhancing data pipelines and facilitating efficient data processing workflows.
4. Neo4j BI Connector for PowerBI. Neo4j provides a Business Intelligence (BI) connector tailored for seamless integration with PowerBI. With this connector, data from Neo4j can be queried using SQL dialect, offering enhanced performance and flexibility. Taking advantage of Neo4j’s graph-native storage format and fast graph traversals, data retrieval demonstrates significantly higher performance compared to traditional relational databases.
5. Microsoft Fabric workload Integration (upcoming). Neo4j will soon be integrated as a native workload for Graph Analytics on the Microsoft Fabric analytics platform. This will enable users to access graph analytics workloads directly from the Microsoft Fabric console, create graph models from OneLake data, analyze graph data, run Graph Data Science Algorithms using Neo4j Bloom as a pluggable UI component, and write back results into OneLake for a seamless end-to-end integration. This integration improves the user experience by blending the capabilities of both platforms.

Showing Neo4j Bloom UI plugged in within Microsoft Fabric console

Redefining What’s Possible in a World Shaped by AI

Delivering enterprise-grade analytics and AI isn’t easy. It requires continual access to clean data from an integrated data analytics and generative AI platform—and that platform needs database technology that can rapidly identify connections within complex datasets while ensuring that GenAI responses meet enterprise standards.

Neo4j enables Microsoft Fabric users to realize the full potential of GenAI and modern analytics—to overcome hallucinations and other GenAI challenges by grounding LLMs with domain-specific data and to deepen business insights by storing information in a graph structure, where intricate relationships within datasets are easy to model and query.

We’re excited about this new strategic partnership with Microsoft because it gives organizations a powerful integrated solution for staying ahead of the GenAI and data analytics curve, both now and for years into the future.

To get started with Neo4j within the Microsoft Fabric ecosystem, explore the GitHub repository for integration resources and get started with Neo4j in the Azure Marketplace today. Unlock the power of graph data to revolutionize your data analytics and drive innovation within your organization.

Welcome to This Week in Neo4j, your weekly fix for news from the world of graph databases!
In this edition, we cover a guide on Cypher debugging, JSON-based agents for LLMs and an Intro to Graphs. Our GraphStuff.FM podcast discusses RAG with graphs and what database to choose for the best outcome.
This week, we started with our Workshop Series. The sessions continue the next week. I hope to see you join us!

COMING UP NEXT WEEK!

• • Conferences: Find us at KubeCon Paris and AI42 Conference on March 19
  • Meetup: Meet us Melborne, AUS on March 20
  • All Neo4j Events: Webinars and More
  • GraphSummit Series: Get Connected With Graphs – Next up: Madrid, ES on April 9
  • Training Series: Hands-On Tutorials, March 14-26
  
  

  FEATURED NODES SPEAKER: Dave Aitel

  
  

  Dave Aitel is well-known in the cybersecurity community for his significant contributions to the field. He has been involved in various aspects of cybersecurity, including offensive security, vulnerability research, and cybersecurity policy.
  Connect with him on LinkedIn.

  In a live stream “Lessons learned from Real-World Graph App Development” he covers Cypher optimisation, the nuances of data importing and practical graph data modeling from his personal history of what he learned since developing with Neo4j. He passes on some valuable lessons to help you navigate the complex landscape of graph app development and achieve optimal results.
  

  CYPHER: Slow Cypher Statements and How to Fix Them

  
  For a couple of weeks now, we have had some reports that some of the quizzes on Neo4j GraphAcademy have been slow. As this is all running on a Neo4j system, Adam Cowley takes this on and explores how to debug a slow Cypher statement.

  GENAI: JSON-based Agents With Ollama & LangChain

  By now, we all have probably recognised that we can significantly enhance the capabilities of LLMs by providing them with additional tools. Tomaz Bratanic shows how to implement a JSON-based LLM agent by using the Mixtral 8x7b as a movie agent to interact with Neo4j through a semantic layer.

  PODCAST: RAG Databases with Johannes Jolkkonen: When to Choose a Graph Database vs Alternative Vector or Relational Stores

  
  In the March episode of GraphStuff.FM Jennifer Reif and Alison Cossette are joined by Johannes Jolkkonen, an independent consultant. With his background in data engineering, he helps companies build their RAG applications; he also creates tutorials on YouTube around RAG, Knowledge Graphs and LLMs.

  GRAPH: Intro to Cypher

  Veronique Gendner shares her slides from an introduction to graph databases and the query language cypher workshop. It is a great starting point for learning what graphs are and understanding Cypher better. Veronique also tells her story on how she discovered graphs and why she likes them.

  TWEET OF THE WEEK: Vasilije

  
  

  A multilayer network in the workshop #llm #graph pic.twitter.com/MsOxU7IX9T

  — Vasilije (@tricalt) March 5, 2024

  
  Don’t forget to share it if you like it! ]]> https://neo4j.com/blog/this-week-in-neo4j-genai-python-springai-php-and-more/ Sat, 09 Mar 2024 16:00:11 +0000 https://neo4j.com/blog/this-week-in-neo4j-graphacademy-algorithms-genai-knowledge-graphs-and-more-copy/ Check out what's new this week in Neo4j, including Superhero dataset, Workspace updates, synthetic data, LLM, streaming graph data, and more.]]>

  Welcome to This Week in Neo4j, your weekly fix for news from the world of graph databases!
  The articles in this edition cover GenAI with JavaScript, a user-friendly Python Library, SpringAI and the Neo4j PHP Driver.
  We are kicking off our Workshop Series next week. I am looking forward to exciting sessions with a broad range of topics.
  
  Don’t forget about O’Reillys Knowledge Graph Course. As readers of our newsletter, you get a free 30-day trial.

  I hope you enjoy this issue,
  Alexander Erdl
  
   

  COMING UP NEXT WEEK!

  • NODES 2023: Watch Recordings
  • Live Stream: Neo4j Live: Entity Resolution and Deduplication with Neo4j and GenAI on March 12
  • Conferences: Find us at AWS Public Sector Symposium Brussels on March 12, Data Innovation Summit, Singapore on March 13 & Data Next Engineering Summit on March 14
  • Meetup: Meet us Delhi, IN and in Bengaluru, IN on March 16
  • All Neo4j Events: Webinars and More
  • GraphSummit Series: Get Connected With Graphs – Next up: Breda, NL on March 13
  • Training Series: Hands-On Tutorials, March 14-26
  
  

  FEATURED NODES SPEAKER: Jacob Marquez

  
  

  Jacob H. Marquez is an insatiable learner and lifelong builder. He is a data scientist/graph engineer by day at an AI startup which aims to reimagine search for shopping.
  Connect with him on LinkedIn.

  In his session at NODES “Using Graphs and Graph Data Science to Unlock the Customer Journey” he shares his Neo4j experience to map customer journeys, revealing crucial behaviours and patterns. He’ll also introduce a framework using Graph Data Science for insight discovery beyond traditional databases’ reach.

  
  
  
   

  GENAI: What is GenAI and can we do it with JavaScript?

  
  Adam Cowley joined Jason Lengstorf, where he showed us how to use langchain.js and build custom GenAI apps. We also just released a new GraphAcademy Course for a Chatbot with TypeScript that goes well with this session.
   

  PYTHON: PyNeoInstance: A User-Friendly Python Library for Neo4j

  PyNeoInstance is a Python package that provides a user-friendly API for submitting Cypher queries to Neo4j by handling tasks such as driver creation, multiprocessing and simple configuration. Alex Gilmore gives you the foundation for your projects in this blog post.
   

  SPRING AI: Spring AI with Neo4j Vector Index

  
  Gerrit Meier takes a closer look at SpringAI, a new module in the pre-release phase that integrates with OpenAI, Bedrock, and Ollama to support embedding creation and chat completion. Combined with the Neo4j Vector Store, this enables customised storage and retrieval of documents and embeddings.
   

  PHP: Neo4j PHP Driver

  The Awesome F/OSS Team outlines in a detailed guide the steps necessary for starting a Neo4j instance locally and integrating the Bolt library into PHP applications. They also add examples of initialising, authenticating and executing queries.
  
  
  

  TWEET OF THE WEEK: Adon

  
  

  Getting ready to publish my launch website for Tract Stack
  
  Doing a clean up of the pre-launch website graph#buildinpublic pic.twitter.com/Cijswnrvjz

  — It’s Adon. (@AtRiskMedia) February 15, 2024

  
  Don’t forget to share it if you like it!
   
   ]]> https://neo4j.com/blog/this-week-in-neo4j-graphacademy-algorithms-genai-knowledge-graphs-and-more/ Sat, 02 Mar 2024 16:00:26 +0000 https://neo4j.com/blog/this-week-in-neo4j-data-journalism-llm-object-mapping-cypher-and-more-copy/ Check out what's new this week in Neo4j, including Superhero dataset, Workspace updates, synthetic data, LLM, streaming graph data, and more.]]>

  Welcome to This Week in Neo4j, your weekly fix for news from the world of graph databases!
  There is a lot of content available for learners in this edition. Besides the updates to GraphAcademy and our Workshops in March, O’Reilly is conducting a Knowledge Graph Course. As readers of our newsletter, you get a free 30-day trial.

  In other news, we have a cool way of looking at Graph Algorithms, learn how to find songs from lyrics and a witty keynote on AI.

  I hope you enjoy this issue,
  Alexander Erdl
  
   

  COMING UP NEXT WEEK!

  • • NODES 2023: Watch Recordings
    • Live Stream: Going Meta – Episode 26 on March 5
    • Conferences: Find us at Women in Tech Day on March 8
    • Meetup: Meet us virtually and in London, UK on March 6 or in Bangkok, TH on March 7
    • All Neo4j Events: Webinars and More
    • GraphSummit Series: Get Connected With Graphs – Next up: Copenhagen, DK on March 7
    • Training Series: Hands-On Tutorials, March 14-26
    
    

  FEATURED NODES SPEAKER: Sebastian Lobentanzer

  
  

  Dr. Sebastian Lobentanzer, a seasoned biomedical researcher, develops innovative solutions for biomedical research; with a passion for accessible data analysis, he creates frameworks for biomedical applications. He actively promotes open-source initiatives, fostering robust and reproducible workflows in the research community.
  Connect with him on LinkedIn.

  In his session at NODES “BioCypher/BioChatter: An Ecosystem for Connecting Knowledge Graphs and LLMs” Sebastian presents an open-source frameworks for knowledge graph creation and LLM interaction, aiming to increase accessibility for the research community. He introduces the BioCypher & BioChatter libraries, which enable the creation of knowledge graphs for various biomedical applications and facilitate interactions between human researchers, LLMs, knowledge graphs, and vector databases.

  
  
  
   

  GRAPHACADEMY: New Courses Available

  
  Our team have released new courses on our Free Online Learning Platform. The latest addition is “Build a Neo4j-backed Chatbot with TypeScript”. We also invite you to our upcoming Training Series in March with various topics from Introduction to Neo4j to RAG App Deployment.

   

  GRAPH ALGORITHMS: Neo4j Graph Algorithms Table of Elements

  Hari Gurumoorthi is an Advanced Data Analytics Specialist at BT Group and published his take on the Neo4j Graph Algorithms visualised as Table of Elements. It is a great overview of the different types of algorithms and looks extremely cool!    

  
  

  GENAI: A Gen AI-Powered Song Finder in Four Lines of Code

  Have you ever wanted to find a song, but you don’t remember what it was called, nor who wrote it? Christoffer Bergman shows us how to build a tool that helps us with that and goes beyond ChatGPT by integrating with Neo4j.

  VIDEO: Lies, Damn Lies, and AIs by Jim Webber

  
  GenerativeAI has taken the world by storm, but it’s not always a reliable helper. It makes up alternative facts, has difficulty with number and logical reasoning, all while exuding the confidence of a dodgy politician! Watch Jim Webber’s closing keynote from JFokus in Stockholm now.

  
  

  TWEET OF THE WEEK: Ed Carron

  
  

  I'm "Launching" the first version of my CompaniesHouseInterface #data investigation tool for analysing business networks, pulling data from UK Companies House and creating an Neo4j graph DB visualisation https://t.co/wtM4p5CkPg pic.twitter.com/VXhn7jO7HD

  — Ed Carron (@EdCarron) February 6, 2024

  
  Don’t forget to share it if you like it!

   
  
    ]]>