On Memory Construction and Retrieval for Personalized Conversational Agents

2025-02-08

AI Summary

To address the challenges of granularity mismatch, low retrieval accuracy, and poor semantic quality in memory construction and retrieval for long-term personalized dialogue, this paper proposes SeCom. First, it introduces a topic-driven dialogue segmentation model to partition dialogues into fine-grained memory units. Second, it integrates LLMLingua-2—a prompt compression technique—to jointly denoise queries and memory entries, thereby enhancing retrieval relevance and semantic fidelity. This work is the first to empirically reveal the dual impact of memory granularity on both retrieval accuracy and semantic quality, and establishes an integrated “segmentation–compression–retrieval” framework that overcomes the limitations of conventional turn-/session-/summary-based modeling. SeCom achieves significant improvements over state-of-the-art (SOTA) methods on LOCOMO and Long-MT-Bench+. Moreover, its dialogue segmentation module sets new SOTA results on DialSeg711, TIAGE, and SuperDialSeg.

Technology Category

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

Application Category

• 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
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

To deliver coherent and personalized experiences in long-term conversations, existing approaches typically perform retrieval augmented response generation by constructing memory banks from conversation history at either the turn-level, session-level, or through summarization techniques. In this paper, we present two key findings: (1) The granularity of memory unit matters: Turn-level, session-level, and summarization-based methods each exhibit limitations in both memory retrieval accuracy and the semantic quality of the retrieved content. (2) Prompt compression methods, such as extit{LLMLingua-2}, can effectively serve as a denoising mechanism, enhancing memory retrieval accuracy across different granularities. Building on these insights, we propose SeCom, a method that constructs a memory bank with topical segments by introducing a conversation Segmentation model, while performing memory retrieval based on Compressed memory units. Experimental results show that SeCom outperforms turn-level, session-level, and several summarization-based methods on long-term conversation benchmarks such as LOCOMO and Long-MT-Bench+. Additionally, the proposed conversation segmentation method demonstrates superior performance on dialogue segmentation datasets such as DialSeg711, TIAGE, and SuperDialSeg.

Problem

• Enhancing memory retrieval accuracy
• Improving semantic quality of retrieved content
• Optimizing long-term conversational coherence

Innovation

• Segmented conversation memory construction
• Compressed memory unit retrieval
• Enhanced long-term dialogue accuracy