arXiv cs.AI· Shuning Zhang, Tenghao Ji, Jingruo Chen, Eryue Xu, Bohao Chu, Xin Yi·· 12 小时前AI 评分34
ConMem：用反事实多场景探测解决 LLM 记忆错位问题
ConMem: Counterfactual Multi-Scenario Probing for Resolving LLM Memory Misalignment
AI 导读
研究者构建 MemAlign 基准（1,079 个多轮对话记忆案例）并提出 ConMem，将记忆保留决策重构为前瞻性反事实因果推断，从任务执行、无关不变性检查与风险敏感边界三方面审计干预效果。实验显示 ConMem 达到 75.16% 准确率、64.15% 平衡准确率和 65.31% Macro-F1，在缓解过度保留偏差的同时不牺牲个性化效用。
正文
Abstract:Long-term memory is critical for autonomous Large Language Model (LLM) agents executing multi-session, long-horizon tasks. However, existing memories often inappropriately store, retain, or retrieve over-generalized, misattributed, or privacy-sensitive information, a problem we term memory misalignment. To study this problem, we construct \textbf{MemAlign}, a benchmark comprising 1,079 curated memory cases grounded in synthetic, validated multi-session conversational contexts covering diverse misalignment taxonomies. We then propose ConMem, an algorithm that reframes retention decisions as prospective counterfactual causal inference. \textbf{ConMem} synthesizes counterfactual interaction scenarios to prospectively audit downstream intervention effects across three aspects: relevant task execution, irrelevant invariance checks, and risk-sensitive boundary shifts. Experiments show that ConMem outperforms other methods, achieving 75.16\% accuracy, 64.15\% balanced accuracy, and a Macro-F1 of 65.31\%, effectively mitigating over-retention biases without sacrificing personalization utility.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.33432 [cs.AI] |
| (or arXiv:2610.33432v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.33432 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Tenghao Ji [view email]
[v1]
Sat, 24 Oct 2026 04:16:37 UTC (634 KB)
来源：arXiv cs.AI · arxiv.org