April 2024 Summaries

Graph-based metadata filtering is being used to improve vector search accuracy in Retrieval-Augmented Generation (RAG) applications. This technique allows users to narrow down their search results according to specific attributes, such as dates or categories. By leveraging the extensive structured information stored in graph databases like Neo4j, sophisticated metadata filters can be executed to precisely refine document selection using structured criteria. The filtering process is combined with vector similarity search to increase accuracy and relevance of search results. In this blog post, LangChain support for metadata filtering in Neo4j is introduced, along with an OpenAI function-calling agent that dynamically generates Cypher statements based on user input and retrieves relevant information from the graph database. The code is available on GitHub, providing a basis for implementing similar solutions in RAG applications.

The text discusses how to add Retrieval Augmented Generation (RAG) capabilities to a GraphQL API using Neo4j. It highlights the Weaviate vector database's native support for RAG and explores how to replicate this functionality in a Neo4j GraphQL project using Langchain.js, an open-source library that enables developers to integrate large language models into their applications. The article provides step-by-step instructions on setting up a basic Neo4j GraphQL project, installing LangChain.js, creating a chain to invoke an LLM, and defining a custom resolver to generate reviews based on user input. It also discusses how to create a generic generate resolver that can be applied to multiple types, such as Movie and Actor, and provides a mapping function to automatically assign resolvers to types that inherit the CanGenerate interface. The article concludes by highlighting the benefits of adding RAG capabilities to GraphQL APIs for dynamic content creation and personalization at scale.

The Neo4j community is actively engaged with the latest developments in graph databases, including the publication of ISO GQL and the Call for Papers for NODES 2024. Various topics such as data modelling for RAG applications, creating knowledge graphs from crime data, and getting started with graphs are covered through articles, links, and events. The community is also invited to join the Neo4j Research panel and participate in upcoming events like Neo4j Live: Knowledge Graph Builder App on May 2 and the GraphSummit Series.

The GQL standard, a new ISO-standardized query language for graph databases, has been published after several years of development. The standard was unanimously approved in March 2024 and is similar to Cypher, the current de facto standard for property graph query languages used by Neo4j. To ensure a smooth transition for users, Neo4j will make its proprietary Cypher implementation GQL-compliant, without offering a separate alternative query language. This means that existing Cypher queries will continue to work, and the language will evolve to better align with SQL and incorporate new features. The GQL standard is more than 600 pages long and includes both mandatory and optional features, which will be supported by Neo4j in the future. Overall, the adoption of the GQL standard is expected to enhance the graph database industry, making it easier for users to switch between different products.

The International Organization for Standardization (ISO) has published GQL, a new database language standard designed specifically for property graphs. This milestone is significant as it marks the first new ISO database language since the introduction of SQL in 1987. GQL, or Graph Query Language, aims to provide a standardized way for users to interact with graph databases, making it easier for companies like Neo4j and Amazon to develop and use graph technologies. The standard has been developed concurrently with Cypher, a popular query language used by Neo4j, and the two languages share many similarities. While some differences exist between GQL and Cypher, the ISO standard provides a foundation for vendors to implement GQL in their products, while also supporting Cypher as an existing implementation of the new standard. This transition is expected to be seamless, with support for both languages available, allowing users to continue using their investments while benefiting from the power of the new standard. The publication of GQL has been eagerly anticipated by the graph community and marks a significant step forward in the development of graph databases.

The GenAI Stack is a pre-built development environment created by Neo4j in collaboration with Docker, LangChain, and Ollama for creating Generative AI (GenAI) applications. It focuses on improving the accuracy, relevance, and provenance of generated responses in Large Language Models (LLMs) through Retrieval Augmented Generation (RAG). The stack includes application containers with Python and LangChain, a database container with Neo4j for vector index and graph search, and an LLM container with Ollama. Key features include information retrieval enhancement using RAG, components such as the database container, functionality to import and embed data, perform queries using vector and graph search, generate new content in the style of existing high-quality data, trust and verification through RAG applications, and customization options like changing parameters of LLMs and creating customized retrievers. The GenAI Stack offers a robust and efficient platform for creating sophisticated GenAI applications, aiding developers in rapid development, testing, and deployment of GenAI applications.

The future of knowledge graphs is a topic that has been explored in various directions. Knowledge graphs have roots in diverse fields, including artificial intelligence, semantic web, and graph theory, and have been built to support various applications over decades. The concept of combining structured representation with the learning capabilities of neural networks is part of a broader trend in AI research that aims to marry symbolic and subsymbolic approaches. Structured search excels in environments where data is well-organized, while semantic search shines in complex, natural language environments. A new approach combines both by storing both structure and meaning in a knowledge graph using Large Language Models (LLMs). This allows for a more robust feedback mechanism and can understand natural language queries regardless of the underlying structure. An experiment on the Movie Graph demonstrates how to implement natural language Q&A using LLMs and Neo4j Graph Database, showcasing promising outcomes from such an approach. Further development will focus on improving efficiency, relevance, and context generation, as well as leveraging metadata and domain knowledge to enhance the system's performance.

RAG, or Retrieval-Augmented Generation, is a machine-learning approach that enhances Large Language Model (LLM) responses by retrieving source information from external data stores to augment generated responses. This technique allows for more accurate and contextual responses, reducing the limitations of standalone LLMs such as hallucinations, lack of explainability, and static training data. By using RAG applications, businesses can provide a personalized experience with domain-specific knowledge, increased accuracy, contextual understanding, explainability, and up-to-date information. Common use cases for RAG include customer support chatbots, business intelligence and analysis, healthcare assistance, legal research, and more.

Entity Resolved Knowledge Graphs (ERKG) are a solution to the problem of duplicate nodes in knowledge graphs, which can dilute their power and accuracy. ERKG uses entity resolution to identify and link data records that refer to the same real-world entities, increasing the accuracy and utility of knowledge graphs. In this tutorial, we build an ERKG using Python and Neo4j, integrating Senzing for entity resolution and leveraging Graph Data Science (GDS) library for graph analytics and visualization. We load three datasets about businesses in Las Vegas, run entity resolution, export results, parse JSON, and connect entities with input records. The resulting ERKG is visualized using PyVis, showing clusters of linked records per entity and the convergence of dataset records achieved through entity resolution.

The ISO committee has published a new standard for Graph Query Language (GQL) in Database Languages, marking graph databases as mainstream alongside traditional databases. This release benefits developers by avoiding the need to learn new technologies while working on multiple projects and products. Various articles cover topics such as GraphRAG on Google Cloud, comparing graph structures, building Knowledge Graphs, and the importance of relationships in graph theory. The Neo4j community is also involved with a research panel, upcoming events, tutorials, and featured community members. Additionally, there are resources for getting started with graphs, including Cypher Indexes and Constraints, a Neo4j Integration for Haystack, and articles on GQL, Knowledge Graphs, and relationships in graph theory.

The text discusses the construction of knowledge graphs from unstructured text data using large language models (LLMs). The approach is based on three steps: extracting nodes and edges from the text, performing entity disambiguation to merge duplicate entities, and importing the data into a Neo4j database. LLMs are used in each step to automate the process, which can be time-consuming if done manually. The code for this project is available on GitHub. However, there are challenges associated with this approach, including unpredictable output formatting from the LLM, speed limitations, and potential lack of accountability. Despite these challenges, the three-step approach enables anyone to build knowledge graphs using LLMs and efficiently analyze large corpora of unstructured data.

ISO has published a new database query language called ISO GQL, which is a peer language to SQL and the first new ISO database language since 1987. The GQL standard is designed for and about graphs, a type of data that excels with connected real- and digital-world systems, and will provide immense value to a landscape where data is increasingly dynamic and interconnected. Graph databases, which use the ISO GQL standard, have gained significant popularity in recent years across various industries, and are increasingly used to solve complex problems. The power of standards will shape industries by providing common methods and patterns that can be applied to solve large-scale problems, and the adoption of graph databases is expected to increase with the growing importance of GenAI and digital transformation. The ISO GQL standard resembles SQL in many ways, but also includes parts specific to graph databases, such as ASCII-art-inspired way of defining patterns. The shortest path to GQL is Cypher, which has already undergone nearly a decade of real-world trial-by-fire maturation and was a major input into the development of the GQL standard.

NODES, a premier developer conference focused on graph-powered applications and contextual AI, is set to return for its sixth year on November 7, 2024. The event will feature live sessions by 110 speakers from around the world, covering topics such as retrieval augmented generation and AI orchestration frameworks. Attendees can expect to learn about the latest GenAI innovations and gain insights into real-world applications of graph technologies. A call for papers is now open until June 15, 2024, with submissions accepted in three session formats: 30-minute talks, 10-minute lightning talks, and 2-hour hands-on workshops. The conference promises to be even better than last year's event, which welcomed thousands of attendees from 30 countries. To get inspiration, visitors can check out the top 10 sessions from NODES 2023 on the conference blog or YouTube playlist.

LangChain has introduced full support for Neo4j Vector Index, a graph database. The library makes it easy to import documents into Neo4j and index them using the vector index. This allows users to perform vector similarity searches on the indexed data. LangChain also supports question-answering workflows using its chat models and retrieval chains, which can be integrated with the Neo4j Vector Index for efficient information retrieval. The library provides a user-friendly interface and customization options for advanced users, making it suitable for various applications such as graph-based research and AI development.

Neo4j is coming up with several upcoming events including a livestream on graph algorithms for data science, conferences in Paris and Texas, and meetups in Austin. The company is also promoting various resources to help developers get started with graphs such as GraphAcademy courses, tutorials, and podcasts featuring experts like Vish Puttagunta. Additionally, Neo4j has released its new book "The Definitive Guide" which offers practical advice on mastering Neo4j enterprise deployments. Furthermore, the company is discussing the integration of PLM-ERP systems using graph models, highlighting their potential for better data management and integration in manufacturing industries. The community is also engaging with each other through tweets and sharing knowledge on various topics including misinformation, fake news, and knowledge graphs.

The Neo4j Professional Services offers a Solution Assessment service to help users maximize their project's potential. This five-day engagement covers various aspects of the project, including requirements and architecture, data model and queries, APIs and data import, GUI, and graph data science (GDS). The assessment includes understanding the project's background and goals, evaluating use cases, assessing graph readiness, focusing on graph development, operations, and providing recommendations and strategic planning. By using this service, users can ensure their Neo4j application meets and exceeds expectations, provides valuable insights into complex data relationships, and aligns with overall strategy.

Graph data models such as parent-child, question-based, and topic-summary can be used to enhance retrieval augmented generation (RAG) applications powered by knowledge graphs. These models provide unique benefits, including vector search capabilities with Neo4j's graph database, which allows for the performance of vector search against the embedding properties of Document nodes to find the most relevant text to return as context to a large language model (LLM). The Parent-Child model further breaks down text chunks into smaller lengths, providing more exact vector search matches and context-rich results. The Questions model links Question nodes to Documents containing useful context for an answer, allowing for two retrieval options: matching LLM-generated questions or real user questions with relationships to highly rated LLM Response nodes. The Topics & Summaries model uses Neo4j's Graph Data Science library to find communities of Document nodes and generate summaries for each community, providing a keyword search in addition to vector search. These models can perform individually or in combination to enhance RAG application capabilities.

The new integrations with Google Cloud and Vertex AI enable organizations to access contextually rich external data to deliver accurate and explainable results in GenAI development. GraphRAG combines retrieval-augmented generation and knowledge graphs, allowing LLMs to reason, infer, and accurately answer questions and execute tasks based on factual information. The integrations simplify the implementation of GraphRAG by providing tools like Gemini models, LangChain, and Neo4j's graph database, enabling developers to quickly create knowledge graphs from unstructured data and ingest real-time data into their applications. This enhances decision-making and user experience across domains and use cases, while reducing hallucinations and improving GenAI accuracy and explainability.

Neo4j has successfully completed its SOC 2 Type 2 report, which attests to its information security policies, procedures, and controls meeting the relevant standards. This certification is granted by an independent third-party auditor and demonstrates Neo4j's commitment to protecting customer data. The company's compliance with HIPAA requirements ensures that sensitive data such as healthcare and financial information is managed securely and in accordance with regulatory standards. Achieving SOC 2 Type 2 compliance provides numerous benefits, including assurance of secure data management, robust security controls, and enhanced customer trust and confidence.

The Neo4j community is hosting several upcoming events and conferences, including QCON in London, Google Cloud Next in Las Vegas, Javaland in Germany, Devnexus in Atlanta, and the AWS Summit in Sydney. Additionally, there will be a meetup in Rome on April 08. The community is also offering various resources to help developers get started with graph technology, such as GraphAcademy courses, webinars, and tutorials. Community members are sharing their expertise and experiences through blog posts, videos, and social media, including a featured community member Saurav Joshi who discussed his project using Neo4j Vector Index and LangChain for a retrieval-augmented generation system. The Neo4j GenAI Ecosystem is also being promoted as a collection of tools and integrations that make it easy to use LLMs with Neo4j, while workshop videos are now available on-demand. Furthermore, Timothy Eastridge has released a book titled "Graph Data Science with Python and Neo4j" which provides a comprehensive guide for enhancing data analysis and insights by integrating Python, Neo4j, and advanced technologies.

This project integrates Neo4j graph databases with LangChain agents, using vector and Cypher chains as tools for effective query processing. The system employs advanced retrieval strategies, enhancing the precision and relevance of information extracted from both vector and graph databases. It features a conversational memory module, ensuring each user interaction is contextually informed. The agents, equipped with these tools, make informed decisions about which retrieval method to use based on the query. This approach optimizes the balance between retrieving specific data and maintaining overall context. The implementation is straightforward, focusing on practical utility and adaptability for different data types. The project aims to improve the efficiency and accuracy of AI-driven data retrieval and processing.