让网页搜索智能体学会判断每一步信息的价值，提升长期决策能力。
ICA: Information-Aware Credit Assignment for Visually Grounded Long-Horizon Information-Seeking Agents
- 将网页内容转为可比的视觉快照，保持布局与多模态信息稳定。
- 通过成功率反推每步信息贡献，为中间步骤分配密集奖励。
- 在多个数据集上显著提升长序列信息搜索性能，适合做智能搜索研究者。
长时序信息搜索智能体的强化学习仍具挑战，因最终奖励仅反映答案正确性，无法指明哪些信息促成成功。该问题在文本生成的网页观测中尤为严重，因解析、截断和摘要常导致内容表示不完整且不稳定。本文提出以证据为核心的框架，将工具获取的信息表示为可识别单元，用于跨轨迹比较。具体地，抓取的网页以渲染快照形式表示，保留版式与多模态内容，作为稳定的观测。在此基础上，提出信息感知信用分配（ICA），一种后处理奖励传播方法，基于回溯轨迹的成功率估算每步的实用度，并为引入高价值信息的中间步骤分配密集奖励。集成至GSPO后，该方法在BrowseComp、GAIA、Xbench-DS和Seal-0上持续提升性能。代码与数据集将公开于https://github.com/pc-inno/ICA_MM_deepsearch。
原文摘要 · Abstract (English)
Long-horizon reinforcement learning for information seeking agents remains difficult because terminal rewards reveal whether the final answer is correct, but not which acquired information enabled it. This difficulty is amplified by text-derived webpage observations, where parsing, truncation, and summarization often produce incomplete and unstable content representations across trajectories. We propose an evidence-centric framework for web agent learning that represents information acquired through tools as identifiable units for comparison across trajectories. In particular, fetched webpages are represented as rendered snapshots, preserving layout and multimodal content as stable content-level observations. Building on these units, we introduce Information-Aware Credit Assignment, a post hoc reward propagation method that estimates turn-level utility scores from rollout success rates and assigns dense rewards to intermediate steps that introduced high-utility information. Integrated with GSPO, our method consistently improves performance on BrowseComp, GAIA, Xbench-DS, and Seal-0. Code and datasets will be released at https://github.com/pc-inno/ICA_MM_deepsearch.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。