GraphRAG-Based Knowledge Extraction with LangChain & Neo4j

Overview

This project implements a Graph Retrieval-Augmented Generation (GraphRAG) system that extracts entities and relationships from text using LangChain and Neo4j. It supports local LLMs via Ollama (Llama3) and cloud-based LLMs like OpenAI’s GPT models.

Unlike traditional RAG, which retrieves document chunks from vector databases, this GraphRAG approach structures knowledge as a graph for enhanced reasoning and retrieval.

Developed in collaboration with CB Laboratory, this project showcases LLMs + knowledge graphs for improved NLP applications.

Features

• GraphRAG Implementation: Enhances retrieval-augmented generation with graph-based reasoning.
• Local & Cloud LLM Support:
  • Ollama (Llama3) for offline processing.
  • OpenAI GPT models via API.
• Entity & Relationship Extraction: Uses LangChain to build a knowledge graph from text.
• Graph Storage & Retrieval: Stores extracted knowledge in Neo4j and retrieves insights via Cypher queries.
• Graph Visualization: Interactive visualization of entity relationships.

Technologies Used

Programming Languages:

• Python – Core development language.
• Cypher – Query language for Neo4j.

Development Environment:

• Google Colab & Local Machines – Supports both cloud and local execution.

Database & Visualization:

• Neo4j – Stores and queries structured knowledge.

Artificial Intelligence:

• Ollama (Llama3) for local processing
• Gemini Pro – LLM used for knowledge extraction
• OpenAI GPT models (cloud API)

Key Python Libraries:

• langchain_google_genai
• langchain_community
• langchain_experiment
• neo4j
• wikipedia
• yfiles_jupyter_graphs
• google.cloud.storage
• pandas

Installation

1. Clone the repository:

   git clone https://github.com/ChristopheCruz/GraphRAG.git
   cd GraphRAG