[论文] SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Coll...

论文概要

研究领域: ML
作者: Yuyao Zhang, Junjie Gao, Zhengxian Wu, Jiaming Fan, Jin Zhang, Shihan Ma, Yao Yao, Weiran Qi, Chuyan Jin, Guiyu Ma, Xingzhong Xu, Kai Yang, Ji-Rong Wen, Zhicheng Dou
发布时间: 2026-07-16
arXiv: 2607.15257

中文摘要

工具集成大语言模型的最新进展使网络搜索成为信息搜索智能体的核心能力。然而，随着交互历史增长，智能体越来越难以追踪任务进度。当搜索尝试未能产生有用的证据时，当前的单智能体和多智能体系统可能陷入重复循环，浪费搜索预算并最终损害最终输出的质量和完整性。我们引入SearchOS，一种系统级多智能体框架，将脆弱、隐式的搜索进度转化为显式、持久和共享的状态。首先，我们将开放域信息搜索表述为基于引用的关系模式补全，智能体在其中发现实体、跨链接表填充属性，并将每个值锚定到源证据。然后我们设计面向搜索的上下文管理（SOCM），它将进化状态外部化为前沿任务、证据图、覆盖图和失败记忆。基于SOCM，SearchOS应用流水线并行调度机制，重叠子智能体的执行，并用针对未解决覆盖缺口的任务持续回填释放的槽位以提高利用率和吞吐量。为调度搜索智能体的执行，SearchOS引入搜索工具中间件 harness，拦截模型和工具交互以记录基于引用的证据，并对停滞或预算耗尽做出反应，还提供可复用的层次化技能系统，包括策略和访问技能，以增强智能体的搜索过程并避免跨运行重复失败的搜索模式。在WideSearch和GISA上，SearchOS在所有评估指标中领先于单智能体和多智能体基线，为稳健的信息搜索协作铺平了道路。

原文摘要

Recent advances in Tool-Integrated Large Language Models have made web search a core capability of information-seeking agents. However, as interaction histories grow, agents increasingly struggle to track task progress. When search attempts fail to yield useful evidence, current single- and multi-agent systems can become trapped in repetitive loops, wasting search budgets and ultimately compromising the quality and completeness of the final output. We introduce SearchOS, a system-level multi-agent framework that turns fragile, implicit search progress into explicit, persistent, and shared state. First, we formulate open-domain information seeking as relational schema completion with grounded citations, where agents discover entities, populate attributes across linked tables, and anchor eac...

自动采集于 2026-07-18

#论文 #arXiv #ML #小凯