LLM Knowledge Graph Builder Demo
In this demonstration, Neo4j showcases its LLM Knowledge Graph Builder, a tool designed to create knowledge graphs from unstructured text for applications like Retrieval-Augmented Generation (GraphRAG). The process begins by connecting to a Neo4j database, which can be set up using Neo4j Aura. Users can upload various types of documents, including PDFs, YouTube transcripts, and Wikipedia pages, to populate the database.
Once connected, users can add documents to the database, such as AI-related content from YouTube, Wikipedia pages on Google DeepMind, and PDFs about AlphaFold releases. After uploading the documents, users select an LLM (Large Language Model) to extract information, configuring what and how to extract based on a predefined schema. This schema includes entities like people, organizations, technologies, and concepts, along with their relationships.
The tool then generates a graph by extracting data from the uploaded documents, which can be visualized to show entities and their interconnections. Users can see the extracted data from each document, such as different Gemini models and Google AI concepts. The visualization also includes cross-connections between entities across different documents.
One of the key features of the Knowledge Graph Builder is its ability to answer questions based on the extracted data. For example, users can ask, “What did DeepMind work on?” The tool utilizes the information from both the graph and the text documents to generate an answer. Additionally, it provides transparency by showing which sections of the documents were used to generate the answer, highlighting specific text chunks and entities involved.
Users can further explore the graph using Neo4j Bloom, a tool for visualizing, editing, and navigating the graph. This allows for a deeper dive into the data, enabling users to understand the relationships and connections within the extracted information.
Overall, Neo4j’s LLM Knowledge Graph Builder offers a powerful way to transform unstructured text into structured knowledge graphs, facilitating advanced data retrieval and analysis through GraphRAG applications.