TAGGRAPH：面向智能体持久历史的标签增强图检索
原标题：TAGGRAPH: Tag-Augmented Graphs for Graph Retrieval of Agent Persistent Histories
AI 摘要
该论文提出 TAGGRAPH 受控评估框架，用统一的 5W 标签表示和 remember()/query() 接口比较不同智能体长期记忆检索配置。在 LongMemEval-S 上 AdaptiveGraph 取得 0.844 MRR，但 BM25 达 0.867、OpenClaw 达 0.880；在 ATANT Core 上局部图遍历优于扩散和 BM25，而 BM25 在压力轮次领先。作者指出图检索与词法检索在不同记忆设置下互有胜负，提取质量会显著制约图配置，且缺失提取标签是前五漏检的常见原因。
正文节选
TAGGRAPH: Tag-Augmented Graphs for Graph Retrieval of Agent Persistent Histories Abstract Long-term memory lets LLM agents recall past interactions and remain consistent across sessions, but memory systems are hard to compare because they often vary in representation, indexing, retrieval, and evaluation. We present a controlled evaluation framework based on shared 5W-style conversational memories. Localized graph configurations traverse a common base graph; AdaptiveGraph adds chronological edges