Large Language Models (LLMs) have gained traction in recommender systems due to their reasoning capabilities and extensive knowledge. However, existing LLM-based agents face challenges such as hallucination and context-length limitations, making them unsuitable for full-ranking recommendation tasks. To address these issues, we propose an agent-based recommendation framework, Memory-based Personalized Recommendation Tool Learning (PRTA), where an LLM acts as a central planner interacting with multiple recommendation models.
In this framework, the LLM agent is responsible for high-level reasoning and personalized tool selection, while traditional recommendation models perform full-ranking scoring, leveraging their scalability in modeling behavioral patterns. To support personalized tool selection, we have designed reflection mechanisms that enable the agent to evaluate and compare tools for each user based on user profiles and candidate ranked lists.
Extensive experiments across three public datasets demonstrate that PRTA outperforms traditional recommendation and LLM-based baselines in improving full-ranking recommendation performance.
Blogger's Review: This research effectively addresses the limitations of LLMs in recommendation systems through a unique agent framework, leveraging the strengths of traditional models and showcasing the future direction of personalized recommendations.