WebCoT：通过在反思、分支与回滚中重构思维链来增强 Web 智能体的推理能力
WebCoT: Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback
AI总结：
本文提出 WebCoT，通过将反思与前瞻、分支和回滚等关键推理技能重构为思维链并进行微调蒸馏，显著提升了 Web 智能体在多个基准上的表现。
摘要：
由大型语言模型（LLMs）驱动的 Web 智能体为下一代人工智能展现出前景，但它们在不确定、动态的 Web 环境中有限的推理能力阻碍了稳健部署。在本文中，我们识别了对有效 Web 智能体至关重要的关键推理技能，即反思与前瞻、分支和回滚，并通过将智能体（推理时）的推理算法重构为思维链理由来整理体现这些能力的轨迹数据。我们在智能体自我改进基准 OpenWebVoyager 中进行了实验，并证明通过简单的微调将显著的推理模式蒸馏到骨干 LLM 中可以大幅提升其性能。我们的方法在多个基准上取得了显著改进，包括 WebVoyager、Mind2web-live 和 SimpleQA（网络搜索），凸显了针对 Web 智能体进行定向推理技能增强的潜力。
英文摘要：
Web agents powered by Large Language Models (LLMs) show promise for next-generation AI, but their limited reasoning in uncertain, dynamic web environments hinders robust deployment. In this paper, we identify key reasoning skills essential for effective web agents, i.e., reflection & lookahead, branching, and rollback, and curate trajectory data that exemplifies these abilities by reconstructing the agent's (inference-time) reasoning algorithms into chain-of-thought rationales. We conduct experiments in the agent self-improving benchmark, OpenWebVoyager, and demonstrate that distilling salient reasoning patterns into the backbone LLM via simple fine-tuning can substantially enhance its performance. Our approach yields significant improvements across multiple benchmarks, including WebVoyager, Mind2web-live, and SimpleQA (web search), highlighting the potential of targeted reasoning skill enhancement for web agents.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
