From Local to Global: A Graph RAG Approach to Query-Focused Summarization
Summary
GraphRAG is a graph-based approach to question answering over private text corpora, aimed at global questions rather than an explicit retrieval task. Standard RAG struggles with global questions, while older query-focused summarization does not scale to a typical RAG index. The method uses a language model to build a graph index in two stages: it derives an entity knowledge graph from the documents, then pregenerates community summaries for groups of related entities. At query time, each community summary yields a partial answer, and those partials are summarized into a final response. On global sensemaking questions over datasets in the 1 million token range, the authors report substantial gains over a conventional RAG baseline in comprehensiveness and diversity. An implementation is at https://github.com/microsoft/graphrag. [S1]
Claims
- id: C1 claim: "GraphRAG builds a graph index by deriving an entity knowledge graph and pregenerating community summaries." source: S1 evidence: "first, to derive an entity knowledge graph from the source documents"
- id: C2 claim: "On global questions over datasets in the 1 million token range, it improves over a conventional RAG baseline." source: S1 evidence: "global sensemaking questions over datasets in the 1 million token range"
- id: C3 claim: "An official implementation is available in a Microsoft GitHub repository." source: S1 evidence: "https://github.com/microsoft/graphrag"
Verification
- Code available: The extract includes https://github.com/microsoft/graphrag, which returned HTTP 200 on 2026-11-12. Not run.
- Reproduced: not reproduced (2026-11-12). Code not run.
- Related work: See 2005-11401-retrieval-augmented-generation, 2310-11511-self-rag, and 2405-14831-hipporag. RAPTOR is in this batch.
- Limits: The reported gains are for a class of global sensemaking questions, not for every local factoid query.
- License: CC BY 4.0 on the abs page (2026-11-12; icon by-4.0.png; http://creativecommons.org/licenses/by/4.0/). Submitted 24 Apr 2024 (v1); last revised v2 (19 Feb 2025). Sources: [S1] https://arxiv.org/abs/2404.16130, v2 (19 Feb 2025). checked_on: 2026-11-12.
When to apply
Use when asking how to answer corpus-level questions with RAG, or how GraphRAG differs from chunk retrieval.
Facts
- id: F1 claim: "GraphRAG builds a knowledge graph and community summaries, then answers from those community-level partials." confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
- id: F2 claim: "The authors report gains over conventional RAG on global questions over 1 million token corpora." confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
- id: F3 claim: "Code is at https://github.com/microsoft/graphrag." confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
Sources
- [S1] arXiv 2404.16130, https://arxiv.org/abs/2404.16130. License: CC BY 4.0. v2 (19 Feb 2025). Accessed 2026-11-12.
License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted). Paper is CC BY 4.0.
License: CC-BY-4.0 · Summary by iinventedlight (AI-assisted). Source: Darren Edge, Ha Trinh, Newman Cheng, and 7 coauthors, From Local to Global: A Graph RAG Approach to Query-Focused Summarization (CC BY 4.0).
Cite this
'iinventedlight (AI-assisted)'. 2026. "From Local to Global: A Graph RAG Approach to Query-Focused Summarization." iinventedlight Library, iil:note:5nxcm1gn, version of 2026-11-12. https://iinventedlight.com/library/notes/2404-16130-graph-rag/. License: CC-BY-4.0.
@misc{iil_note_5nxcm1gn,
author = {{'iinventedlight (AI-assisted)'}},
title = {{From Local to Global: A Graph RAG Approach to Query-Focused Summarization}},
howpublished = {iinventedlight Library},
year = {2026},
date = {2026-11-12},
url = {https://iinventedlight.com/library/notes/2404-16130-graph-rag/},
note = {Permanent ID iil:note:5nxcm1gn. License: CC-BY-4.0.},
}
Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.
For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2404-16130-graph-rag.md.