https://howardatri.github.io/tags/rl/Recent content in RL on howardatriHugozh-cnTue, 04 Aug 2026 13:15:29 +0800https://howardatri.github.io/posts/why-cot-falls-short/Tue, 04 Aug 2026 13:15:29 +0800https://howardatri.github.io/posts/why-cot-falls-short/<p><strong>论文笔记:《Rethinking Reasoning in Document Ranking: Why Chain-of-Thought Falls Short》</strong> arxiv25</p> <p><strong>作者:</strong> Xuan Lu, Haohang Huang, Rui Meng, Yaohui Jin, Wenjun Zeng, Xiaoyu Shen <strong>机构:</strong> Shanghai Jiao Tong University, Eastern Institute of Technology, Ningbo</p> <h3 id="1-研究背景">1. 研究背景</h3> <h4 id="11-问题定义">1.1 问题定义</h4> <ul> <li> <p>文档重排序 (Document Reranking) 是 IR 系统的关键组件，用于精炼检索结果，提升排序质量 。</p> </li> <li> <p>受大型推理模型 (LRMs, 如 DeepSeek-R1, o1) 的启发，近期研究开始将显式的“思维链”(CoT) 推理引入 LLM Reranker 。 via supervised fine-tuning Weller et al. (2025); Ji et al. (2025); Yang et al. (2025) or using reinforcement learning Zhang et al. (2025); Zhuang et al. (2025); Liu et al. (2025).</p>https://howardatri.github.io/posts/search-r1/Sat, 01 Aug 2026 22:02:34 +0800https://howardatri.github.io/posts/search-r1/search-r1论文解读