June 2024 Summaries

This week in Neo4j brings news and updates from the world of graph databases, focusing on GraphRAG/Knowledge Graphs with hands-on videos, detailed challenges and recommendations, chatbots, and innovative GenAI Apps. Ashleigh Faith, a PhD expert in Advanced Semantics, is featured as a community member and has been confirmed to speak at NODES 2024 about adding statement verification to Knowledge Graph processes. The Neo4j newsletter also highlights the importance of knowledge graphs for advanced generative AI applications, such as LLMs grounding off data in KGs.

mongo2neo4j is a tool that enables developers to transfer their MongoDB data into a native graph, specifically Neo4j. This allows for the visualization and exploration of the data graph behind every MEAN Stack application. SemSpect, a no-code graph exploration and visualization tool, can be used to gain insights into business data without requiring any query language skills or knowledge of the underlying graph model. By using mongo2neo4j and SemSpect together, developers can establish a configurable workflow to connect the dots in their MongoDB database and find data flaws or answer relevant BI questions. The workflow is free, but may require switching to enterprise versions for larger datasets or professionalization.

The author of the text describes a technique called "Mix and Batch" for fast, parallel relationship loading in Neo4j. The technique involves dividing start node IDs and end node IDs into non-overlapping sets, combining them to create partition codes, laying out these codes into a table, gathering the codes into batches with unique column and row IDs per batch, partitioning the data by the partition code within each batch, and loading one batch at a time. The author tested this approach with large datasets in Neo4j AuraDS environment using Apache Spark and achieved significant speed improvements, up to two to three times faster than traditional serial loading methods. This technique can be applied to streaming data, but requires further investigation and experimentation.

Knowledge graphs are being increasingly used to unlock the full potential of data assets for AI applications, with many firms exploring or planning to use them this year. Firms have achieved early wins with AI in areas such as customer experience and cost-saving efforts, but knowledge graphs can provide a deeper understanding of data, enabling more strategic decisions. Knowledge graphs are powerful frameworks that capture relationships between entities from a real-world domain using a graph structure, which formats data as an interconnected network. They excel at connecting data across disparate systems and data sources, making them well-suited for applications requiring real-time analytics. The arrival of generative AI has brought a demand for knowledge-based reasoning, with 48% of FIMA survey respondents planning to use techniques to improve responses generated by AI. Knowledge graphs are recognized as the industry standard for enterprise-ready AI, and many leading analyst firms agree that they power a new generation of AI capable of providing a deep understanding of data.

The Property Graph Index is a modular component of LlamaIndex that allows for customizable graph construction and retrieval. It uses a property graph data structure, where nodes have labels and optional node properties. The index is designed to be flexible and adaptable, with various graph constructors and retrievers available. In this blog post, the authors explore how to implement entity deduplication and design custom retrieval methods to enhance GraphRAG accuracy. They demonstrate how to construct a knowledge graph using schema-guided extraction, perform entity deduplication using text embedding similarity and word distance, and implement a custom retriever that uses entity detection and vector context retrieval. The authors also discuss the importance of entity disambiguation and provide guidance on how to customize the retriever for specific use cases. Overall, the Property Graph Index offers a powerful toolkit for building knowledge graphs and improving graph-related accuracy.

The Neo4j community is abuzz with recent updates and exciting developments in the world of graph databases. The LLM Fundamentals GraphAcademy Course has been updated to reflect the latest Langchain release v0.2, introducing new features such as LCEL (Langchain Expression Language) and using Neo4j as a conversation memory store. Nobel Prize Outreach is exploring the use of Knowledge Graphs to uncover connections between laureates for storytelling and interactive visualizations, while also delving into the basics behind GraphRAG and its applications in predicting outcomes, such as the French Open Roland Garros. The community is also coming together for various events, including Neo4j Live, AI Engineer World Fair, and NODES 2024, where experts will share their insights on graph-based technologies and innovative solutions.

The development of a new ISO/IEC standard language called Graph Query Language (GQL) marks a significant milestone in the database industry. GQL was first introduced as Cypher, which evolved into openCypher and finally transitioned into the draft ISO standard GQL in 2019. The latest major revision of SQL, SQL:2023, has been published, introducing new features for property graph queries over data stored in tables and more support for JSON data types. This represents a pivotal development for advancing data analytics and management in the era of big data, GenAI, and beyond. GQL addresses the need for sophisticated queries that reflect the increasingly complex and connected nature of data in the real world, providing more intuitive and powerful ways to visualize and interrogate data relationships while storing and processing data more efficiently.

The Neo4j LLM Knowledge Graph Builder is an innovative online application that transforms unstructured text into a knowledge graph with no code and no Cypher, providing a "magical text-to-graph" experience. It uses ML models (LLMs) to extract entities and relationships from PDFs, web pages, YouTube videos, and other sources, and constructs a graph format using Neo4j's capabilities. The application provides a seamless user interface for data ingestion, entity recognition, graph construction, and visualization of the generated knowledge graph. With its Retrieval-Augmented Generation (RAG) chatbot, users can interact with their data in a natural language query and receive explainable insights into their data. The LLM Knowledge Graph Builder is part of Neo4j's GraphRAG Ecosystem Tools and offers friction-free access to this powerful technology for GenAI applications.

The Neo4j GraphRAG Ecosystem Tools are a suite of open-source tools designed to help developers create GenAI applications grounded in knowledge graphs. These tools, including the Knowledge Graph Builder and NeoConverse, make it easy to get started with GenAI development by providing a structured way to work with unstructured text data and query knowledge graphs using natural language. The tools integrate seamlessly with popular open-source GenAI ecosystem libraries for Python, JavaScript, Java, and .Net, allowing developers to build custom implementations or use pre-built templates. With the GraphRAG tools, developers can create a knowledge graph from unstructured text, retrieve relevant information for generative tasks via vector and graph search, and visualize and question the ingested data to improve response quality and explainability. The tools are designed to accelerate app development and adoption, providing a solid foundation for building GenAI applications that can provide accurate and reliable responses.

Neo4j Runway is a Python library developed with Jason Booth and Dan Bukowski at Neo4j, which provides tools for abstracting communication with OpenAI to run discovery on data and generate a data model. The library also includes modules for generating ingestion code and loading data into a Neo4j instance. To install, simply run the pip command `pip install neo4j-runway`. Runway uses Graphviz to visualize data models and requires a Neo4j instance to use fully. It can automatically write ingestion code for your Neo4j graph and load your data into a Neo4j instance. The library is still in beta, but it simplifies the user experience of understanding how relational data fits into a graph data model and allows users to explore quickly. Runway's key features include data discovery, graph data modeling, code generation, data ingestion, and visualization using Graphviz.

The Neo4j Runway Python library is a tool for exploring and ingesting relational data into a Neo4j graph database. It provides a simplified user experience by abstracting communication with OpenAI to run discovery on the data and generate a data model, as well as tools for generating ingestion code and loading data into a Neo4j instance. The library relies on large language models (LLMs) from OpenAI to provide valuable insights from the data. It supports two code generation options: LOAD CSV and PyIngest YAML configuration, both of which generate necessary Cypher code to create constraints and load CSV data. The library also provides a built-in ingestion function via a modified PyIngest file. Runway is still in beta and encourages feedback and bug reporting.

The Neo4j community is actively engaging in various events and conferences, including NODES 2024, where a call for papers closes on June 15. Upcoming events include livestreams, meetups, and webinars, showcasing the versatility of graph databases like Neo4j. Featured community member Jonathan Looi will share insights into using Neo4j for cyber threat detection within cloud environments at NODES 2024. Additionally, Llamaindex introduces a new Property Graph Index feature, enabling more flexible knowledge graph construction and querying. The Graph Data Science plugin is also highlighted, allowing users to uncover malicious threat actor behavior in cybersecurity applications. Furthermore, Semantic Search techniques using Neo4j GDS can enhance document retrieval in RAG applications, while an introduction to Neo4j provides a basic understanding of its capabilities for data modeling and querying. Entity resolution with knowledge graphs improves data accuracy and clarity, and Qdrant's integration with APOC procedures brings advanced vector search capabilities to graph database applications.

The text discusses the adoption of the OWASP Application Security Verification Standard (ASVS) in a Neo4j engineering organization. The team found that using ASVS was underutilized compared to the OWASP Top 10, and they wanted to make it more accessible. They used graph data and chatbots to create a tool that helps developers identify relevant security requirements for their features. The tool uses a knowledge graph of ASVS requirements, calculates vector embeddings for each property, and retrieves relevant requirements based on the input feature description. The team tested the tool with an example feature description and obtained a list of semantically similar requirements, which were then passed to an LLM for further analysis. The output was a sensible summary of security considerations for the feature, without hallucinations. The tool is designed to be easy to use and can drive engagement with ASVS and the overall security culture of the organization.

The LlamaIndex Neo4j Integration Starter Kit is a comprehensive repository that enables developers to store, index, and query documents with unparalleled efficiency and flexibility using the power of natural language processing (NLP) and graph databases. The kit seamlessly combines the robustness of Neo4j with the efficient indexing and querying capabilities of LlamaIndex, empowering users to create conversational AI assistants, document retrieval systems, and other NLP applications. With a Python notebook and FastAPI application demonstrating how to integrate these technologies, the starter kit provides everything needed to unlock the true potential of NLP and graph databases. By following simple steps, developers can get started with the starter kit and embark on their journey to build amazing NLP applications with LlamaIndex and Neo4j.

The GenAI Starter Kit is a pre-packaged solution for building generative AI (GenAI) applications with Spring AI in Java. It provides a simplified process by including starter kit projects with key technologies, such as Neo4j and OpenAI models. The project uses Neo4j to store structured data and unstructured text data with vector embeddings, allowing for similarity searches and retrieval queries. The application model consists of domain classes that map entities to the database model, and a controller class that handles user requests and generates responses using the Neo4jVectorStore and OpenAiChatClient beans. The starter kit includes examples and resources for getting started with GenAI and beyond.

The introduction of a validated partner solution between Databricks and Neo4j enables enterprises to combine structured and unstructured data, discover hidden patterns across billions of data connections, and enhance contextual understanding within their data. This integration allows developers to ingest data from Databricks into Neo4j for analysis in a graph structure, uncovering hidden patterns and generating deeper insights through pre-built graph algorithms and the Cypher query language. The solution also unlocks the potential of GenAI with knowledge graphs and retrieval-augmented generation (RAG), improving LLM accuracy, relevance, and transparency, and accelerating GenAI app development by seamlessly incorporating GraphRAG capabilities into projects.

The Mahabharata chatbot is a technological marvel that brings the epic Indian tale to life through a context-rich and intuitive interface. It leverages the power of graph databases like Neo4j, using a revolutionary approach called GraphRAG that combines retrieval-based models with Large Language Models (LLMs). This allows the chatbot to generate more nuanced and factually accurate responses, making it an engaging experience for users. The chatbot also employs a tool called Neo4j LLM Knowledge Graph Builder, which creates a powerful knowledge graph by analyzing vast amounts of information about the Mahabharata. This web of interconnected information captures the essence of the epic, enabling the chatbot to provide insightful and contextually rich responses.

This week's edition of This Week in Neo4j features Leann Chen as a guest on the podcast, discussing knowledge graphs and their applications. The newsletter also covers integration testing, including a new approach using Testcontainers, and provides an overview of how to create a Knowledge Graph in just 100 lines of code. Additionally, it highlights upcoming events, including livestreams, conferences, meetups, and webinars, as well as features on community members and their projects, such as Vraj Routu's work with GenAI.

The article explores the differences between graph search systems and vector search systems in retrieval-augmented generation (RAG) systems. It uses a financial report RAG example to compare the two types of answer outputs, optimizing depth and breadth through graph structures, and discovering why combining graph and vector search is the future of RAG. Graph databases like Neo4j store data and relationships, enabling explicit querying and reducing hallucinations. Knowledge graphs provide a comprehensive view of all relevant information, enhancing reasoning and extraction capabilities. By defining a schema, users can create well-scoped knowledge graphs efficiently, ensuring that the graph accurately reflects the relationships inherent in the raw data. The article also explores how combining graph and vector search using graph structures can augment retrieval, providing more complete answers in both depth and breadth, while creating a semantically consistent, accurate, and deterministic way to perform information retrieval.

The generative AI market has seen significant growth, leading to increased demand for knowledge graphs. Knowledge graphs are semantically rich data models that store entities, relationships, and an organizing principle to capture meta-information about core concepts. Two popular approaches to building knowledge graphs are RDF (Resource Description Framework) and property graphs. RDF is a W3C standard for data exchange on the Web, but it has limitations, such as not being able to model multiple distinct relationships of the same type. Property graphs, on the other hand, offer flexibility and scalability, allowing for fast querying and traversal, and are designed specifically as a database model. They have advantages over RDF, including simplicity, detailed data storage, interoperability, and standards compliance. The property graph model has been successfully implemented in various industries, such as NASA, Basecamp Research, and Novo Nordisk, to solve complex problems, reduce costs, and achieve breakthroughs. When choosing between a knowledge graph approach, it is recommended to use the property graph model by default and layer in organizing principles from the RDF world when needed.

The integration of Neo4j's Graph Data Science (GDS) capabilities with the Snowflake AI Data Cloud enables users to perform graph analytics directly within Snowflake SQL, without requiring expertise in graph analysis. This allows for a serverless, isolated, and flexible environment for graph analytics, which can be used for various applications such as predictive analytics, fraud detection, and generative AI. The integration eliminates the need for complex data migration jobs between separate ML or database platforms, reducing time to value and allowing organizations to get more analytics projects into production faster. With Neo4j's extensive library of graph algorithms, including similarity, pathfinding, community detection, and many others, users can leverage advanced insights in various industries, such as banking, healthcare, supply chain, and more. The integration is available for preview and early access on Snowflake Marketplace, with general availability later this year.

Neo4j has launched a new GraphAcademy course on Data Importing, and a livestream series called GoingMeta which covered topics such as knowledge graphs, semantics, and more. The company is also hosting various events including NODES 2024, where developers can submit their graph-powered projects to share with the community. A new GraphAcademy Course for Data Importing has been released, and tips have been provided for submitting a talk at NODES 2024. Additionally, the company's GenAI team is exploring the use of graphs in LLM conversations, while promoting the adoption of graph databases through various initiatives, including connecting with community members such as Sören Klein.