GraphRAG
Retrieval over a knowledge graph built from the corpus.
Overview
GraphRAG extracts entities and relationships from a document collection, builds a graph, and answers questions by traversing it and summarising communities of related material. It solves the problem that flat chunk retrieval cannot answer questions which require connecting facts scattered across many documents. It exists because some questions are about the shape of a corpus rather than any one passage in it.
Technology Graph
How GraphRAG connects to the rest of the stack.
Uses and depends on
- Embedding concept Techniques GraphRAG uses Embedding High ✓ source
- LLM concept Models GraphRAG uses LLM High ✓ source
- RAG technique Techniques GraphRAG based on RAG High ✓ source