﻿Microsoft GraphRAG paper knowledge graph / 微软GraphRAG论文知识图谱
Source: From Local to Global: A GraphRAG Approach to Query-Focused Summarization, arXiv:2404.16130v2, 19 Feb 2025.
The graph is a Codex-curated analysis of the supplied paper, not a replication of its experiments.
中文：website/zh/knowledge-graph.html；English: website/en/knowledge-graph.html.
Both HTML files embed the complete public graph and work offline.
Evidence quotes retain PDF page and character offsets. Chinese interpretations are not original quotations.
Graph: 80 entities, 105 semantic relations, 48 evidence spans.
Neo4j: imported into the independent dataset microsoft_graphrag_paper_v2_20261021.
Actual GDS: PageRank, Louvain, WCC; four relationship-chain queries returned paths.
Actual Microsoft GraphRAG: create_communities / hierarchical Leiden only.
No model API calls, Microsoft community reports, embeddings or live retrieval queries.
No credentials, .env, runtime binaries, cache or unrelated datasets are included.
The public website reads this snapshot; it does not connect to a D: drive or local Neo4j.
Neo4j assertions and evidence are separate records. Evidence checks verify extraction consistency, not paper factual accuracy.
Reproduction scripts require the existing local runtimes noted in the scripts; do not run build_paper_graph.py after analysis unless recomputing both analyses.
