AgentCF++: Memory-enhanced LLM-based Agents for Popularity-aware Cross-domain Recommendations

TL;DR

AgentCF++ addresses the problem of cross-domain user behavior simulation in LLM-based recommender systems by introducing a dual-layer memory for users (domain-separated and domain-fused memories) and a two-step fusion mechanism. It also models popularity effects via interest groups and a group-shared memory, allowing information to influence similar users while limiting cross-domain noise. Empirical results on Amazon-derived cross-domain datasets show that AgentCF++ outperforms AgentCF and other baselines, with ablations confirming the value of both dual-layer memory and interest-group sharing. The work advances practical cross-domain simulation with explicit popularity modeling, offering a scalable approach for more realistic user-agent behavior and recommendations.

Abstract

LLM-based user agents, which simulate user interaction behavior, are emerging as a promising approach to enhancing recommender systems. In real-world scenarios, users' interactions often exhibit cross-domain characteristics and are influenced by others. However, the memory design in current methods causes user agents to introduce significant irrelevant information during decision-making in cross-domain scenarios and makes them unable to recognize the influence of other users' interactions, such as popularity factors. To tackle this issue, we propose a dual-layer memory architecture combined with a two-step fusion mechanism. This design avoids irrelevant information during decision-making while ensuring effective integration of cross-domain preferences. We also introduce the concepts of interest groups and group-shared memory to better capture the influence of popularity factors on users with similar interests. Comprehensive experiments validate the effectiveness of AgentCF++. Our code is available at https://github.com/jhliu0807/AgentCF-plus.

Figures (1)

• Figure 1: (a) The memory propagation process in AgentCF. (b) An example illustrating AgentCF's limitations in modeling user behavior influenced by popularity factors. (c) Illustration of why modeling popularity factors is necessary for accurately simulating user behavior. (d) Overview of the proposed AgentCF++ model, highlighting its improvements over AgentCF.