Enhancing Cross-Domain Recommendations with Memory-Optimized LLM-Based User Agents

2025-02-19

AI Summary

Existing LLM-based user agents for cross-domain recommendation suffer from low memory efficiency, information redundancy, and the neglect of social influences—particularly item popularity. To address these issues, we propose AgentCF++, a novel framework featuring a two-tier memory architecture (individual memory + shared interest group memory) and a two-step preference fusion mechanism. Crucially, AgentCF++ is the first to explicitly model popularity-driven social influence within LLM-based user behavior modeling. The shared interest group memory dynamically captures cross-domain popularity trends by clustering users with similar preferences, while the two-tier structure effectively decouples personalized signals from social signals, mitigating domain shift. Extensive experiments on multiple cross-domain benchmark datasets demonstrate that AgentCF++ significantly outperforms state-of-the-art baselines in recommendation accuracy, robustness, and interpretability.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Application Domains: Humanities & Computational Social Science
• Multiagent Systems: Modeling other Agents

Application Category

• User Modeling, Personalization and Recommendation: Social recommender systems and personalization for the social Web
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Vertical and domain-specific search

Abstract

Large Language Model (LLM)-based user agents have emerged as a powerful tool for improving recommender systems by simulating user interactions. However, existing methods struggle with cross-domain scenarios due to inefficient memory structures, leading to irrelevant information retention and failure to account for social influence factors such as popularity. To address these limitations, we introduce AgentCF++, a novel framework featuring a dual-layer memory architecture and a two-step fusion mechanism to filter domain-specific preferences effectively. Additionally, we propose interest groups with shared memory, allowing the model to capture the impact of popularity trends on users with similar interests. Through extensive experiments on multiple cross-domain datasets, AgentCF++ demonstrates superior performance over baseline models, highlighting its effectiveness in refining user behavior simulation for recommender systems. Our code is available at https://anonymous.4open.science/r/AgentCF-plus.

Problem

• Addresses inefficiency in cross-domain recommender systems.
• Introduces dual-layer memory for domain-specific preference filtering.
• Captures social influence factors through shared memory groups.

Innovation

• LLM-based user agents
• dual-layer memory architecture
• interest groups with shared memory