arXiv cs.AI· Miaobo Hu, Shuhao Hu, BoKun Wang, Yina Sa, Xin Wang, Xiaobo Guo, Daren Zha·· 12 小时前AI 评分26
ECPO：面向验证器审计候选排序的证据耦合策略优化
ECPO: Evidence-Coupled Policy Optimization for Verifier-Audited Candidate Ranking
AI 导读
研究提出证据耦合策略优化（ECPO），让策略在输出 Top-K 排序的同时生成与候选关联、基于原文片段的证据包，并结合列表级效用、证书有效性与证据循环奖励。在 MAVEN-ERE 和 RAMS 上，ECPO 相比匹配对照组提升了验证器审计可追溯性、排序盲恢复与人工对比支持，且在留出评估器、其他 8B/14B 骨干模型和 HotpotQA 答案排序接口下依然成立。
正文 · 原文
Abstract:Ranking systems used for decision support should expose evidence that can be audited independently of the model's ranking scores. We study a closed-roster setting in which a policy jointly emits a Top-$K$ ranking and candidate-linked, span-grounded evidence bundles. Evidence-Coupled Policy Optimization (ECPO) combines listwise utility, certificate validity, and an evidence-cycle reward that removes answer-bearing fields before deterministic recovery. On MAVEN-ERE and RAMS, ECPO improves verifier-audited traceability, rank-blind recovery, and human comparative support relative to matched controls in the tested provenance settings. The gains persist under held-out evaluators, other 8B/14B backbones, and a HotpotQA answer-ranking interface. Ordinary NDCG and runtime remain explicit trade-offs.
| Comments: | 42 pages, 1 figure. Revised manuscript; updated authorship and corresponding author information |
| Subjects: | Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2605.21993 [cs.AI] |
| (or arXiv:2605.21993v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2605.21993 arXiv-issued DOI via DataCite |
Submission history
From: Miaobo Hu [view email]
[v1]
Thu, 21 May 2026 04:42:34 UTC (1,768 KB)
[v2]
Fri, 16 Oct 2026 10:38:10 UTC (3,322 KB)
来源：arXiv cs.AI · arxiv.org