GraphRAG, introduced by Microsoft Research in 2024, is a next-generation approach designed for the "global" questions that classic chunk-based RAG handles poorly. It extracts entities and relationships from documents, builds a knowledge graph, summarizes that graph via community-detection algorithms, and brings the relevant summaries into context at query time. It outperforms standard Vector Database-driven RAG on synthesis questions like "what themes run across this report set?" The build cost is high — it needs an LLM pipeline to construct the graph — so it's not the right fit for every RAG use case.
MEVZU N°124ISTANBULYEAR I — VOL. III
Glossary · Advanced · 2024
GraphRAG
A RAG variant that extracts a knowledge graph from documents to capture the bigger picture.
- EN — English term
- GraphRAG
- TR — Turkish term
- GraphRAG