GraphRail：搜索智能体中间步骤级验证方法
Where Search Agents Go Off the Rails: Graph-Grounded Step-Level Verification
AI 导读
论文提出 GraphRail，将问题需求、证据与推理表示为结构化声明并相互比对，无需生成式评审或标准答案即可逐步检查搜索智能体推理是否偏离。该方法在三个多跳 QA 基准和三个推理框架上提升错误召回率，并在随机抽取下保持稳定，其诊断结果还可引导在线推理，实现中间错误的可追溯检测与修正。
正文 · 原文
Abstract:Search-augmented reasoning systems can reach correct answers while making errors in intermediate steps. Reliable auditing must check whether evidence $D$ covers the question $Q$, reasoning $T$ addresses $Q$, and $T$ is grounded in $D$. We introduce GraphRail, which represents requirements, evidence, and reasoning as structured claims and compares them without a generative judge or gold answer. Across three multi-hop QA benchmarks and three reasoning frameworks, GraphRail improves error recall over generative verifiers while remaining stable under stochastic extraction. Its diagnoses can also guide online reasoning, providing a traceable way to detect and correct intermediate errors.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.06662 [cs.AI] |
| (or arXiv:2611.06662v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.06662 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Hyeseo Jeon [view email]
[v1]
Wed, 4 Nov 2026 15:37:36 UTC (317 KB)
来源：arXiv cs.AI · arxiv.org