In Prospect and Retrospect: Reflective Memory Management for Long-term Personalized Dialogue Agents

2025-03-11

AI Summary

Large language models (LLMs) struggle to consistently model and retrieve user-specific information over extended personalized dialogues. Method: We propose a Prospective/Retrospective Reflection mechanism—a dual-path introspection framework—that overcomes two key limitations of existing external memory systems: rigid memory granularity and static retrieval policies. Our approach enables dynamic, multi-granular memory construction across utterances, turns, and sessions; introduces an online reinforcement learning framework for retrieval optimization grounded in LLM-generated citation evidence; and integrates an external memory bank with multi-granular dynamic summarization. Contribution/Results: Evaluated on LongMemEval, our method achieves over 10% absolute accuracy improvement. It demonstrates consistent performance gains across diverse metrics and scenarios, marking the first unified modeling of elastic memory granularity and context-adaptive retrieval strategies.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large Language Models (LLMs) have made significant progress in open-ended dialogue, yet their inability to retain and retrieve relevant information from long-term interactions limits their effectiveness in applications requiring sustained personalization. External memory mechanisms have been proposed to address this limitation, enabling LLMs to maintain conversational continuity. However, existing approaches struggle with two key challenges. First, rigid memory granularity fails to capture the natural semantic structure of conversations, leading to fragmented and incomplete representations. Second, fixed retrieval mechanisms cannot adapt to diverse dialogue contexts and user interaction patterns. In this work, we propose Reflective Memory Management (RMM), a novel mechanism for long-term dialogue agents, integrating forward- and backward-looking reflections: (1) Prospective Reflection, which dynamically summarizes interactions across granularities-utterances, turns, and sessions-into a personalized memory bank for effective future retrieval, and (2) Retrospective Reflection, which iteratively refines the retrieval in an online reinforcement learning (RL) manner based on LLMs' cited evidence. Experiments show that RMM demonstrates consistent improvement across various metrics and benchmarks. For example, RMM shows more than 10% accuracy improvement over the baseline without memory management on the LongMemEval dataset.

Problem

• LLMs struggle with long-term personalized dialogue retention.
• Existing memory mechanisms fail in semantic structure capture.
• Fixed retrieval mechanisms lack adaptability to diverse contexts.

Innovation

• Dynamic summarization across conversation granularities
• Online reinforcement learning for memory retrieval
• Integration of forward and backward reflection mechanisms