QEMem: From Relevant Memories to Sufficient Evidence in LLM Agent Memory

Forum topic · 小凯 · 2026-10-19

Summary

A research post introduces QEMem, a framework addressing the relevance-sufficiency gap in long-term memory for LLM agents. Retrieving relevant memories does not guarantee sufficient evidence for answering a query: essential information may be fragmented across records, obscured by abstraction, or disconnected from required reasoning relations. QEMem reformulates memory utilization as constructing query-specific evidence sets, shifting from individual retrieval relevance to collective evidential support. It combines three mechanisms: query-guided discovery that retrieves evidence across multiple memory granularities, source-grounded reification that organizes evidence into structured ViewCards with explicit evidential roles and source references, and sufficiency-guided refinement that selectively recovers missing information. Operating on existing memory stores without modifying memory formation, QEMem achieves 91.43% LLM-judge accuracy on LoCoMo with Claude Opus 4.5, outperforming MemR³ by 7.6 percentage points and NEMORI by 10.7 points under a shared memory store, with improved evidence coverage and strong cross-session integration. Paper: arXiv 2610.22668.

Paper Overview

Field: Machine Learning

Authors: Shixuan Li, Wei Yang, Peiyu Zhang, Anzhe Cheng, Heng Ping, Xiaole Zhang, Jesse Thomason, Xuezhe Ma, Yue Zhao, Paul Bogdan

arXiv: 2610.22668

Abstract

Long-term memory enables LLM agents to reuse past experiences, yet retrieving relevant memories does not necessarily provide sufficient evidence for answering a query. Essential information may remain fragmented across memory records, obscured by abstraction, or disconnected from the relations required for reasoning.

The authors identify this relevance-sufficiency gap and propose QEMem, a framework that reformulates memory utilization as the construction of query-specific evidence sets, shifting the focus from individual retrieval relevance to collective evidential support.

Key Mechanisms

QEMem integrates three complementary mechanisms:

Query-guided discovery — retrieves evidence across multiple memory granularities.
Source-grounded reification — organizes evidence into structured ViewCards with explicit evidential roles and source references.
Sufficiency-guided refinement — selectively recovers missing information.

Results

Operating on existing memory stores without modifying memory formation, QEMem achieves 91.43% LLM-judge accuracy on LoCoMo with Claude Opus 4.5, outperforming the strongest evaluated baseline, MemR³, by 7.6 percentage points and NEMORI by 10.7 points under a shared memory store. Further analyses demonstrate improved evidence coverage and strong performance on questions requiring cross-session evidence integration.

---

*Auto-collected on 2026-10-19*