arXiv cs.CL· Yikai Zhu, Kunfeng Chen, Qihuang Zhong, Juhua Liu, Bo Du·· 3 天前AI 评分27
ConRAG：面向多跳问答的共识驱动证据流框架
ConRAG: Consensus-Driven Evidence Flow for Multi-Hop Question Answering
AI 导读
ConRAG 是一个共识驱动证据流框架，通过多视角共识排序筛选相关证据，并用槽位机制在依赖的推理步骤间传递中间信息，以提升 LLM 的多跳问答表现。在三个多跳 QA 基准上，ConRAG 在 Gemma-4-31B-it 上平均分超出最强基线 5.8 分，端到端运行时间和 token 消耗分别降低 75.9% 和 67.3%。
正文 · 原文
Abstract:Retrieval-augmented generation (RAG) has emerged as a promising paradigm for enhancing large language models (LLMs) on multi-hop question answering (QA), which requires reasoning over evidence from multiple documents. However, current multi-hop RAG approaches still struggle with evidence selection and information transfer, i.e., retrieving irrelevant passages or failing to pass intermediate information to subsequent reasoning steps, thereby hindering coherent multi-hop reasoning. To this end, we propose ConRAG, a consensus-driven evidence flow framework that effectively boosts LLM performance on complex multi-hop QA. In short, the core of ConRAG is to select relevant evidence via consensus-driven multi-view ranking, and leverage a slot-based mechanism to transfer intermediate information between dependent reasoning steps. Extensive experiments across three multi-hop QA benchmarks show that ConRAG surpasses the baselines by a clear margin. Encouragingly, on Gemma-4-31B-it, ConRAG outperforms the strongest baseline by 5.8 points in average score, while reducing total end-to-end runtime and token consumption by 75.9% and 67.3%, respectively.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2605.28093 [cs.CL] |
| (or arXiv:2605.28093v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.28093 arXiv-issued DOI via DataCite |
Submission history
From: Yikai Zhu [view email]
[v1]
Wed, 27 May 2026 07:51:46 UTC (1,243 KB)
[v2]
Tue, 9 Jun 2026 08:12:08 UTC (1,243 KB)
[v3]
Wed, 14 Oct 2026 16:22:16 UTC (1,230 KB)
来源：arXiv cs.CL · arxiv.org