arXiv cs.AI· Haotian Liu, Yingjian Chen, Xiaohang Xu, Lang Huang, Yusuke Iwasawa, Yutaka Matsuo, Irene Li·· 3 天前AI 评分27
DDRG：通过采样推理轨迹间的图对齐实现 LLM 推理的细粒度纠错
Beyond Answer Voting: Fine-Grained Correction of LLM Reasoning through Graph Alignment across Sampled Traces
AI 导读
DDRG 是一个基于冻结推理轨迹库的选择性纠错框架，通过构建有依据的声明图并做图对齐，把局部冲突交给不看答案的验证器，仅当验证器支持的驳斥能触及多数答案的支持路径时才触发一次定向修复，否则返回原始 SC 投票结果。
正文 · 原文
Abstract:Large language models increasingly tackle tasks that require multi-step reasoning, where errors in intermediate claims can affect the final answer. Self-consistency (SC) improves reasoning by voting over sampled final answers, but its vote leaves intermediate disagreements unresolved. To address this problem, we introduce DDRG (Disagreement-Driven Verification Guided by Reasoning Graphs), a framework for selective correction from a frozen bank of reasoning traces. Our insight is that graph alignment can expose claim-level disagreements and provide correction evidence without expanding the sampled trace bank. DDRG builds source-grounded claim graphs and sends localized conflicts to an answer-blind verifier. A verifier-supported refutation that reaches the majority answer's supporting path triggers one focused repair; otherwise, DDRG returns the exact SC vote. Experimental results across five benchmarks show that native zero-shot DDRG improves macro-average accuracy over direct answering by 8.65 and 15.83 percentage points for Qwen2.5-7B and Llama3.1-8B. It also exceeds CoT, the evaluated GoT and GoV question-answering adaptations, and SC on this macro metric. DDRG links local verification to selective correction without enlarging the initial candidate bank.
| Comments: | 19 pages, 5 figures, 4 tables. Under review |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.18606 [cs.AI] |
| (or arXiv:2610.18606v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.18606 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Haotian Liu [view email]
[v1]
Tue, 13 Oct 2026 15:58:02 UTC (421 KB)
来源：arXiv cs.AI · arxiv.org