arXiv cs.CL· Xiaomeng Hu, Jiaqi Hu, Hao Chen, Qi Zhang, Zhanming Shen, Wentao Ye, Junbo Zhao·· 2 天前AI 评分29
针对 LLM 推理幻觉的 SSC-GRPO：步级自洽性组相对策略优化
Reasoning Error from Known Fact: Step-Level Self-Consistency Group Relative Policy Optimization for LLMs
AI 导读
研究者提出 Step-level Self-Consistency Group Relative Policy Optimization（SSC-GRPO），通过计算多次 rollout 中单个推理步骤的自洽性分数，为推理轨迹分配步级奖励。
正文 · 原文
Abstract:With the rapid advancement of large language models (LLMs), modern systems not only possess strong foundational capabilities and extensive knowledge, but can also solve complex problems via long, multi-step reasoning. However, as reasoning traces become longer, LLMs may produce a substantial amount of hallucinated content during the reasoning process, which is often difficult to detect. In this work, we conduct a fine-grained analysis of hallucinations arising in LLM reasoning and find that the reasoning traces are particularly prone to Context-Sensitive Factual Hallucinations: cases where the model actually has the relevant knowledge, yet makes factual errors due to contextual interference during reasoning. To address this issue, we propose Step-level Self-Consistency Group Relative Policy Optimization (SSC-GRPO), which assigns step-level rewards to reasoning traces by computing self-consistency scores of individual steps across multiple rollouts. Compared with prior methods, SSC-GRPO achieves state-of-the-art performance on both mathematical reasoning benchmarks and hallucination leaderboards. Our results offer a new perspective for detecting and mitigating hallucinations in the reasoning process of large language models.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2607.18915 [cs.CL] |
| (or arXiv:2607.18915v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.18915 arXiv-issued DOI via DataCite |
Submission history
From: Xiaomeng Hu [view email]
[v1]
Tue, 21 Jul 2026 09:56:58 UTC (544 KB)
[v2]
Thu, 15 Oct 2026 08:25:16 UTC (578 KB)
来源：arXiv cs.CL · arxiv.org