幸运回忆：面向LLM持久连贯性的本体驱动记忆生命周期管理
Fortunate Recall: Ontology-Driven Memory Lifecycle Management for Persistent Coherence in LLMs
浏览论文内容
中文总结 AI 辅助
提出本体驱动的记忆生命周期管理方法，按行为类型分类个人事实并应用差异化策略，在多个基准上提升检索正确率并大幅降低虚构。
中文摘要 AI 辅助
当前的LLM记忆系统将所有个人事实等同对待，导致存储无界增长而检索精度下降。核心挑战在于生命周期管理：根据每个事实的行为类型，哪些记忆应持久保留、哪些应被替换、以及以何种速率进行。幸运回忆（Fortunate Recall, FR）是一个可组合的策略层，将个人事实分类为10+1行为本体，并应用类别特定的生命周期策略（差异化时间衰减、槽位键替代、事件时间有效性和类别感知检索路由），作为LLM提取元数据上的确定性函数。FR-Bank是我们与基础设施无关的实现，在LifecycleBench（一个包含516个问题的时间消歧新基准）上达到76.9%的通过率，领先于Mem0、A-MEM、Memory-R1和MemoryOS（61%至70.5%），并在规范Wu等人评判协议下于完整LongMemEval-S上达到75.2%，表明生命周期策略对标准检索无可测量成本。预注册消融实验定位了增益来源：用三个通用生命周期原语替换类型化层后，正确性统计上无显著变化（-1.7个百分点，95%置信区间[-6.0, +2.7]），因此通用生命周期元数据承载正确性优势，而行为本体承载校准能力，将下游虚构减半（12.0%对24.2%，p<0.001）。端到端地，FR-Bank在已回答查询上将虚构从Mem0的45.1%降至22.4%，在所有查询上从32.2%降至13.0%，同时正确回答更多查询（31.2%对18.6%）；排名在开放权重Kimi K2.5上复现。该分解迁移至独立构建的BEAM基准：在280个问题上正确率46.8%，优于Mem0的32.9%，本体收益集中于矛盾消解，并在约七个策略簇处饱和。本体、基准和代码已发布。
英文摘要
Current LLM memory systems treat all personal facts identically, so stores grow without bound while retrieval precision degrades. The core challenge is lifecycle management: which memories should persist, which should be replaced, and at what rate, conditioned on the behavioral type of each fact. Fortunate Recall (FR) is a composable policy layer that classifies personal facts into a 10+1 behavioral ontology and applies category-specific lifecycle policies (differential temporal decay, slot-key supersession, event-time validity, and category-aware retrieval routing) as deterministic functions over LLM-extracted metadata. FR-Bank, our infrastructure-independent implementation, reaches a 76.9% pass rate on LifecycleBench, a new 516-question temporal-disambiguation benchmark, ahead of Mem0, A-MEM, Memory-R1, and MemoryOS (61% to 70.5%), and 75.2% on the full LongMemEval-S under the canonical Wu et al. judge protocol, so lifecycle policies impose no measurable cost on standard retrieval. A pre-registered ablation locates the gains: replacing the typed layer with three generic lifecycle primitives leaves correctness statistically unchanged (-1.7pp, 95% CI [-6.0, +2.7]), so the generic lifecycle metadata carries the correctness advantage, while the behavioral ontology carries calibration, halving downstream confabulation (12.0% vs 24.2%, p<0.001). End-to-end, FR-Bank cuts confabulation from Mem0's 45.1% to 22.4% over answered queries and from 32.2% to 13.0% over all queries while answering more of them correctly (31.2% vs 18.6%); the ranking replicates on the open-weight Kimi K2.5. The decomposition transfers to BEAM, an independently built benchmark: 46.8% correct vs Mem0's 32.9% over 280 questions, with the ontology's benefit concentrated in contradiction resolution and saturating near seven policy clusters. The ontology, benchmark, and code are released.
发表机构
- Bilkent University（比尔肯大学）
机构由 AI 辅助整理，请以论文原文为准。
