From Retrieval to Reconstruction: Constructing Evolvable Cognitive Memory for Long-Term Dialogue

2026-10-08

AI Summary

This study addresses the limitations of existing Retrieval-Augmented Generation (RAG) frameworks that treat memory as passive storage, struggle to distinguish subjective from objective information, and fail to correlate cross-session evidence. To overcome these challenges, we propose CogMem, a cognitive memory architecture that incrementally constructs memory based on a PEC²F graph. This method employs Claim nodes to decouple subjective assertions from objective facts, and introduces a temporal-scope conflict reconciliation mechanism alongside a rule-driven retrieval controller. Query contexts are reconstructed through LLM-based intent parsing combined with deterministic graph operators such as anchoring and traversal, while knowledge consolidation and semantic folding probes enable reliable reasoning in long-term dialogues. Experiments demonstrate that CogMem significantly improves performance on multi-hop, temporal, and knowledge-updating tasks across the LoCoMo and LongMemEval benchmarks, validating the complementary contributions of its individual modules.

Abstract

Large Language Models (LLMs) serving as long-term dialogue agents require memory systems that support reliable reasoning over extended interactions. However, existing Retrieval-Augmented Generation (RAG) frameworks typically treat memory as passive storage, making it difficult to distinguish source-attributed beliefs from unattributed event/fact records and to connect evidence dispersed across sessions. We introduce CogMem, a cognitive memory architecture based on the PEC$^2$F (Person-Event-Concept-Claim-Fact) graph schema. Dedicated Claim nodes preserve the source and target of subjective statements, while Fact and Event nodes represent semantic and episodic knowledge. Dialogue turns are incrementally converted into provenance-aware graph records, consolidated into higher-level facts, and reconciled into temporally scoped Claim views when the same source provides conflicting updates. For retrieval, a rule-based controller driven by LLM intent parsing composes four deterministic graph operators---anchoring, traversal, intersection, and evidence grounding---to reconstruct query-relevant context. Experiments on LoCoMo and LongMemEval show strong performance, especially on multi-hop, temporal, and knowledge-update tasks. Ablations and a semantic-collapse probe support complementary contributions from epistemic separation, consolidation, and agentic retrieval. Code: https://github.com/Silent-Rain02/CogMem.

Problem

• Long-term Dialogue
• Cognitive Memory
• Retrieval-Augmented Generation
• Large Language Models
• Memory Systems

Innovation

• Cognitive Memory
• Knowledge Graph
• Long-Term Dialogue
• Provenance-Aware Retrieval
• Memory Consolidation