MemoCRS: Memory-enhanced Sequential Conversational Recommender Systems with Large Language Models

TL;DR

This paper tackles the problem of modeling user preference continuity across multiple dialogue sessions in conversational recommender systems. It proposes MemoCRS, a memory-enhanced framework that uses a per-user entity-based memory (UM) to compress historical preferences and a general memory (GM) to capture collaborative knowledge and reasoning guidelines, enabling LLMs to deliver more accurate and personalized recommendations. The approach combines memory retrieval, external expert knowledge, and LLM-driven reasoning to handle both typical and cold-start users, and it is evaluated on Chinese TGReDial and English ReDial datasets, where MemoCRS achieves substantial gains over strong baselines in both recommendation and dialogue quality. The findings demonstrate that memory-augmented LLMs can effectively model sequential user preferences, with general memory particularly beneficial for cold-start scenarios, suggesting practical improvements for scalable, cross-domain CRSs.

Abstract

Conversational recommender systems (CRSs) aim to capture user preferences and provide personalized recommendations through multi-round natural language dialogues. However, most existing CRS models mainly focus on dialogue comprehension and preferences mining from the current dialogue session, overlooking user preferences in historical dialogue sessions. The preferences embedded in the user's historical dialogue sessions and the current session exhibit continuity and sequentiality, and we refer to CRSs with this characteristic as sequential CRSs. In this work, we leverage memory-enhanced LLMs to model the preference continuity, primarily focusing on addressing two key issues: (1) redundancy and noise in historical dialogue sessions, and (2) the cold-start users problem. To this end, we propose a Memory-enhanced Conversational Recommender System Framework with Large Language Models (dubbed MemoCRS) consisting of user-specific memory and general memory. User-specific memory is tailored to each user for their personalized interests and implemented by an entity-based memory bank to refine preferences and retrieve relevant memory, thereby reducing the redundancy and noise of historical sessions. The general memory, encapsulating collaborative knowledge and reasoning guidelines, can provide shared knowledge for users, especially cold-start users. With the two kinds of memory, LLMs are empowered to deliver more precise and tailored recommendations for each user. Extensive experiments on both Chinese and English datasets demonstrate the effectiveness of MemoCRS.

Figures (4)

• Figure 1: An example of leveraging the user preference continuity to assist in conversational recommendations. The yellow and green rectangles denote the utterances from the user and the system, respectively.
• Figure 2: The overall framework of MemoCRS.
• Figure 3: Comparison between different kinds of memory.
• Figure 4: Performance comparison on cold and warm users.