Memory clearance boosts dialog agent task success by removing misleading data
MeClear: Cooperative Game-Theoretic Attribution and Risk-Aware Memory Clearance for Long-Horizon LLM Agents
Artificial IntelligenceSoftware Engineering
Summary
Long conversations with AI helpers need memory to remember what you said earlier, but sometimes that memory includes confusing or wrong information. The researchers designed MeClear, a system that figures out which memories actually hurt the AI's ability to do its job right. It uses clever math to see how different pieces of information work together and then clears out only the harmful parts without losing important data. Tests showed this approach helps the AI succeed much more often compared to older methods.
Large Language Modelsmemory systemstask knowledgeretrieval mechanismsShapley attributionLeave One Out screeningcontext clearancedownstream utilitytask recoverylong dialogue
Authors
Boyu Yang, Jiazheng Sun, Zilong Lu, Zhi Qiu, Xin Peng, Jun Zheng
Abstract
Long horizon Large Language Model (LLM) agents rely on external memory systems to preserve user preferences and task knowledge across extended interactions. Conventional retrieval mechanisms optimize semantic compatibility rather than downstream utility, frequently introducing outdated, misleading, or conflicting evidence into the active context. We present MeClear, a task conditioned memory clearance framework that identifies memories featuring negative downstream utility through cooperative attribution and selectively suppresses them from agent execution. MeClear combines Leave One Out screening with sampled cooperative Shapley attribution to distribute utility across interacting evidence, effectively resolving redundant conflict masking where single removal evaluations fail. Utilizing attribution rankings, MeClear executes a query scoped minimal clearance strategy over a nested filtration, verifying task recovery on the cleared context without permanently altering the persistent memory bank. Comprehensive experimental evaluations across ten long dialogue memory pools demonstrate that MeClear achieves a target recall of 85.9% and an overall task recovery rate of 82.3%, representing a 25.5 percentage point improvement over Leave One Out (LOO) baselines.