NLP & Language · advanced · concept 148 of 311 · Updated Nov 4, 2026
GraphRAG
GraphRAG combines retrieval with a knowledge graph. Plain RAG pulls passages that sound similar to a question; GraphRAG can also follow relationships between entities and summarize across a whole collection. That helps with questions needing several connected facts, at the price of building and updating the graph, so it should be tested against simpler retrieval first.
L7 · LLM and generative AI engineerDevelopersTech leadersProduct leadersQA and testingSEO and content
Key terms
Entity extractionGraph traversalCommunity summariesMulti-hop questionsGraph upkeep
Learn these first
Where you meet it in the real world
Investigations across documents, research synthesis, enterprise knowledge assistants
Guides and articles
GraphRAG: discovery on narrative private data ↗
Microsoft Research
microsoft/graphrag ↗
Microsoft
From Local to Global: A Graph RAG Approach ↗
Edge and others