This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
To overcome the limitations of existing Retrieval-Augmented Generation (RAG)-based recommender systems, this paper proposes the Agentic RAG (ARAG) framework, which integrates a multi-agent collaboration mechanism. ARAG utilizes four specialized LLM-based agents: a user understanding agent, a natural language inference (NLI) agent, a context summarization agent, and an item ranking agent, to understand users' long-term and short-term behaviors. Experimental results show that ARAG outperforms existing RAG and state-of-the-art baseline models by up to 42.1% on NDCG@5 and up to 35.5% on Hit@5. The effectiveness of each agent is analyzed through ablation studies.
Takeaways, Limitations
•
Takeaways:
◦
Improving the Performance of Personalized Recommendation Systems through a Multi-Agent-Based RAG Framework
◦
A novel approach that effectively considers users' long-term and short-term behaviors is proposed.
◦
A New Direction in LLM-Based Personalized Recommendation System Research
◦
Suggesting directions for system improvement through analysis of the effectiveness of each component of ARAG.
•
Limitations:
◦
Only experimental results for a specific dataset are presented, so further research is needed to determine generalizability.
◦
Lack of detailed description of the interaction and collaboration mechanisms between agents.
◦
Lack of consideration for actual system construction and operation
◦
Lack of analysis of performance changes according to the type and size of LLM used.
