DOTRAG: Retrieval-Time Reasoning Along Paths
The paper introduces DotRAG, a new framework for Graph Retrieval-Augmented Generation that reformulates retrieval as a reasoning process. This approach aims to improve performance on complex multi-hop tasks by generating query-conditioned constraints that guide graph exploration. DotRAG has demonstrated state-of-the-art performance on benchmarks like MetaQA and UltraDomain.
- ▪DotRAG is a training-free framework that enhances the retrieve-then-reason paradigm in GraphRAG.
- ▪The framework generates constraints that help in exploring relevant relational paths while pruning irrelevant areas.
- ▪DotRAG has achieved state-of-the-art performance on multiple benchmarks, particularly in multi-hop tasks.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Information Retrieval arXiv:2605.18760 (cs) [Submitted on 6 Apr 2026] Title:DOTRAG: Retrieval-Time Reasoning Along Paths Authors:Larnell Moore, Naihao Deng, Rada Mihalcea, Farnaz Jahanbakhsh View a PDF of the paper titled DOTRAG: Retrieval-Time Reasoning Along Paths, by Larnell Moore and 3 other authors View PDF HTML (experimental) Abstract:Graph Retrieval-Augmented Generation (GraphRAG) is dominated by a retrieve-then-reason paradigm, where context is retrieved using heuristics and then reasoned over. Such methods struggle to adapt to the query-specific logic required for complex multi-hop tasks, often accumulating irrelevant context or missing correct relational paths.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
