ETAMem Decouples Heterogeneous Memory for Long-Term Conversational Agents
October 28, 2026
ETAMem addresses information entanglement in LLM memory by parsing atomic units and organizing them into three hierarchical views: Event, Affect, and Time. This multi-view structuring prevents semantic similarity from overriding temporal or emotional associations in long-term interactions.
HOW THIS AFFECTS YOU
●
builderYou can improve personalization in long-term agents by decoupling emotional and temporal data from pure semantics.
●
researcherYou can explore new ways to structure heterogeneous conversational data beyond simple semantic retrieval.