ArcRAG enables construction-agnostic graph retrieval
Original title: ArcRAG: Agentic Compositional Retrieval over Knowledge Graphs
Query-local graph interface induction allows fixed retriever to work on differently built knowledge graphs.
GraphRAG answers questions by retrieving over a corpus-derived knowledge graph. Retrieval strategies often couple to their construction pipeline. ArcRAG induces a query-local graph at inference time and plans multi-hop traversals as Cypher. Experiments on MuSiQue, HotpotQA and TriviaQA show the retriever stays effective when the graph construction changes, though absolute accuracy depends on construction.
- Query-time interface induction works across graph families
- Retrieval need not co-design with graph construction
- Ablations show retry and pass budget impact
- Construction shapes absolute accuracy
- Fixed configuration works on both typed and open graphs
From the original article
1 min read at arxiv.orgGraph-based retrieval-augmented generation (GraphRAG) answers questions by retrieving over a knowledge graph built from a corpus. Yet retrieval strategies are designed around the structure their own construction pipeline produces, and whether they survive a change of construction is rarely tested.
By Jyotin Goel · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.