执行溯源何时有助于智能体记忆检索？

AI 摘要

该论文研究语言智能体在上下文窗口受限时，如何从执行历史中检索完整证据。作者提出将智能体记忆检索形式化为「预算内证据补全」问题，并构建 Execution Provenance Graph Memory（EPGM），用零初始化残差 R-GCN 在工具参数与输出的溯源单元上精炼稠密检索分数。在 ISETrace 的 2,000 条查询上，溯源单元相比固定 512 token 窗口将 Full Support@2048 提升 19.07 点，图传播在候选固定时额外提升 4.55 点，且增益集中在证据跨多个执行事件的情况。

正文节选

When Does Execution Provenance Help Agent Memory Retrieval? Abstract. A language agent’s execution history can exceed its context window, requiring its memory system to retrieve complete supporting evidence under a hard token budget. Evidence may span multiple execution events, yet conventional retrievers use fixed token windows and fixed- metrics that reward individual fragments without showing whether the complete evidence set fits in context. Smaller windows reduce irrelevant text but scatter

发布时间：2026-09-23 12:00

抓取时间：2026-09-23 12:09

来源机构：arXiv