// the find
DEEP-PolyU/Awesome-GraphRAG
Awesome-GraphRAG: A curated list of resources (surveys, papers, benchmarks, and opensource projects) on graph-based retrieval-augmented generation.
A literature survey repo tracking GraphRAG papers — the subset of RAG that routes retrieval through knowledge graphs or graph-structured indexes rather than flat vector search. Backed by a PolyU research group that also publishes in this space (LinearRAG, GraphRAG-Bench). Aimed at researchers and practitioners who want a map of the field before building something or writing a paper.
The taxonomy is actually useful — splitting Knowledge Organization, Retrieval, and Integration gives you a mental model, not just a flat dump of links. The benchmark table is the best part: it lists dataset, task type, and repo links in one place, saving an hour of hunting. The group actively maintains it with their own accepted papers (ICLR'26, ACL'26, KDD'26), so it reflects what's winning at top venues right now, not just what was trending in 2024. Having GraphRAG-Bench as a concrete evaluation harness rather than just paper claims is genuinely useful if you're comparing approaches.
No code, no runnable examples — it's entirely links to PDFs. The open-source project section is a shallow list of GitHub badges with one-line descriptions that tells you nothing you couldn't get from the repo names. Several papers appear multiple times across sections without any note explaining why (RAPTOR, Think-on-Graph, Medical Graph RAG), which makes the taxonomy feel leaky. The survey paper itself is an arXiv preprint that's been sitting unreviewed since January 2025, so the framing of the whole repo rests on work that hasn't been independently vetted.