GraphRAG generates a knowledge graph from unstructured documents by extracting entities and their relationships using large language models. This structured graph enables more accurate and explainable retrieval-augmented question answering.
The Microsoft GraphRAG implementation used in this project is available at : GraphRAG Microsoft
To use it:
- Follow Microsoft’s official setup guide:
Link : Get Started with GraphRAG - To visualize the knowledge graph using Gephi, do the following:
- First, run the script in the GraphRAG_Microsoft folder after the graphRAG generation:
It aims at formating the data so the links are visible on Gephi.
python generate_graphml_with_relationships.py
- Then, follow Microsoft’s visualization guide:
Link : GraphRAG Visualization Guide
- First, run the script in the GraphRAG_Microsoft folder after the graphRAG generation:
Once you have set up your workspace variables (follow the Microsoft tutorial I linked in point 1.), you will have access to ragtest/settings.yaml. Feel free to modify it as you want for your applications.