This project demonstrates the integration of a Neo4j graph database, a vector-based search system, and large language models (LLMs) to create a robust system for hybrid search and question answering. It combines structured graph-based data and unstructured document retrieval, offering precise and context-rich responses to queries.
- Hybrid Search: Combines Neo4j graph queries and vector-based semantic search for more accurate results.
- Entity Extraction: Extracts entities (e.g., people, organizations) from text using an LLM-based pipeline.
- Knowledge Graph Construction: Converts documents into graph entities and relationships using a graph transformer.
- Unstructured Data Retrieval: Leverages embeddings for semantic similarity search on text documents.
- Question Condensing: Handles follow-up questions by rephrasing them into standalone queries.
- End-to-End Query Handling: Processes both structured and unstructured data sources to generate concise, relevant answers.
Ensure you have the following installed and configured:
-
Python Libraries:
langchain-corelangchain-openailangchain-communityneo4jsentence-transformersgoogle-colab(for cloud setup)yfiles-jupyter-graphs(for graph visualization)
-
Environment:
- Neo4j database instance with credentials.
- OpenAI API key for LLM integration.
-
Tools:
- A Python environment (e.g., Jupyter Notebook, Colab) with widget support.
- Clone this repository:
git clone https://github.com/your-username/your-repo.git cd your-repo