arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2604.11544cs.CLcs.AI
收藏

时间并非标签:连续相位旋转用于时间知识图谱与智能记忆

Time is Not a Label: Continuous Phase Rotation for Temporal Knowledge Graphs and Agentic Memory

University of Edinburgh(爱丁堡大学);LIGHTSPEED;University of St Andrews(圣安德鲁斯大学);Independent Researcher(独立研究员)

Weixian Waylon Li, Jiaxin Zhang, Xianan Jim Yang, Tiejun Ma, Yiwen Guo

AI总结:

本文提出RoMem模块,通过连续相位旋转技术解决时间信息在知识图谱中的表示问题,提升时间推理和记忆保持性能,实现状态-of-the-art结果。

摘要:

结构化记忆表示,如知识图谱,是自主代理和其他长生命周期系统的核心。然而,现有方法大多将时间视为离散元数据,要么按时间排序(埋藏旧但永久的知识),要么简单覆盖过时事实,或在每次摄入步骤中需要昂贵的LLM调用,使其无法区分持久事实与演变事实。为了解决这一问题,我们引入RoMem,一种适用于智能记忆和更广泛结构化记忆系统的时序知识图谱模块。一个预训练的语义速度门将每个关系的文本嵌入映射到一个波动性分数,通过学习数据中演变关系(如“总统”)应快速旋转,而持久关系(如“出生在”)应保持稳定。结合连续相位旋转,这使几何阴影成为可能:过时事实在复杂向量空间中被旋转出相位,因此时间正确事实自然优于矛盾,而无需删除。在时间知识图谱完成任务中,RoMem在ICEWS05-15上取得最先进的结果(72.6 MRR)。应用于智能记忆时,它在时间推理(MultiTQ)上交付2-3倍的MRR和回答准确性,主导混合基准(LoCoMo),保持静态记忆无降级(DMR-MSC),并能泛化到未见的金融领域(FinTMMBench).

英文摘要:

Structured memory representations such as knowledge graphs are central to autonomous agents and other long-lived systems. However, most existing approaches model time as discrete metadata, either sorting by recency (burying old-yet-permanent knowledge), simply overwriting outdated facts, or requiring an expensive LLM call at every ingestion step, leaving them unable to distinguish persistent facts from evolving ones. To address this, we introduce RoMem, a drop-in temporal knowledge graph module for structured memory systems, applicable to agentic memory and beyond. A pretrained Semantic Speed Gate maps each relation's text embedding to a volatility score, learning from data that evolving relations (e.g., "president of") should rotate fast while persistent ones (e.g., "born in") should remain stable. Combined with continuous phase rotation, this enables geometric shadowing: obsolete facts are rotated out of phase in complex vector space, so temporally correct facts naturally outrank contradictions without deletion. On temporal knowledge graph completion, RoMem achieves state-of-the-art results on ICEWS05-15 (72.6 MRR). Applied to agentic memory, it delivers 2-3x MRR and answer accuracy on temporal reasoning (MultiTQ), dominates hybrid benchmark (LoCoMo), preserves static memory with zero degradation (DMR-MSC), and generalises zero-shot to unseen financial domains (FinTMMBench).

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。

↑