GraphRAG

by Microsoft

System Card

OrganizationMicrosoft

Released2024-03

Architecturegraph-rag / Entity-community hierarchical graph

DetailsData pipeline that extracts entities, relationships, and claims via LLM, then clusters them into hierarchical communities (Leiden). Global queries summarize across community reports; local queries traverse entity neighborhoods with vector fallback.

Parameters—

Domainrag-retrievalknowledge-graph

Open SourceYes

PaperView Paper

WebsiteVisit

graphknowledge-graphcommunity-detectionmicrosoft-researchmit

Capability Profile

Benchmark Scores

6 of 14 benchmarks

Long-Context Retrieval

2/5

72.462p

NIAH

no data

LooGLE

no data

603p

∞Bench

no data

Multi-Turn Recall

0/2

LoCoMo

no data

MemoryBank

no data

Cross-Session Memory

1/1

71.829p

Multi-Hop QA

2/3

BABILong

no data

7468p

71.551p

Agent Task Memory

0/1

AgentBench-Mem

no data

Personalization

0/1

PerLTQA

no data

Factuality / Grounding

1/1

73.181p

Sources:GraphRAG paper (arXiv:2404.16130); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)GraphRAG paper (arXiv:2404.16130); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)GraphRAG paper (arXiv:2404.16130); evaluated on RAGAS: Automated Evaluation of Retrieval-Augmented Generation (Exploding Gradients, 2309)GraphRAG paper (arXiv:2404.16130); evaluated on LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding (Tsinghua KEG, 2308)GraphRAG paper (arXiv:2404.16130); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)GraphRAG paper (arXiv:2404.16130); evaluated on RULER: What's the Real Context Size of Your Long-Context Language Models (NVIDIA, 2404)