cs.AI提交于 2026-10-19已译

ECPO: 用于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

Miaobo Hu · Shuhao Hu · BoKun Wang · Yina Sa · Xin Wang · Xiaobo Guo · et al.

中文摘要

用于决策支持的排序系统应展示可独立于模型分数进行审计的证据。本研究探讨封闭名单设置，策略联合输出 Top-K 排序及与候选项关联的跨度接地证据包。证据耦合策略优化（ECPO）结合列表效用、证书有效性和证据循环奖励，在确定性恢复前移除含答案字段。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、无排名恢复及人工比较支持。增益在留出评估器、其他 8B/14B 骨干及 HotpotQA 答案排序界面中持续存在。普通 NDCG 和运行时仍是明确的权衡。

关键要点

01解决排序系统证据无法独立于模型分数进行审计的问题。
02提出 ECPO，结合列表效用、证书有效性和证据循环奖励优化策略。
03在 MAVEN-ERE 和 RAMS 上提升验证器审计可追溯性及无排名恢复能力。
04增益在留出评估器、不同骨干模型及 HotpotQA 界面中保持稳健。
05指出普通 NDCG 指标和运行时性能是明确的权衡代价。

解读

尚无解读。

原始英文摘要

arXiv:2605.21993v2 Announce Type: replace 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.