Adaptive Conversation Recommendation Systems: Leveraging Large Language Models and Knowledge Graphs

This paper introduces a novel conversation recommendation system that leverages large language models (LLMs) and knowledge graphs (KGs) to enhance recommendation accuracy and adaptability. Our system dynamically analyzes user queries using LLMs, decomposing them into meaningful graph nodes within a reference KG like YAGO. These nodes then form the basis of personalized user-centric KGs, representing individual preferences. Node weights are adjusted based on ongoing conversations, capturing evolving user intent. To ensure relevance, decay rate factor is implemented to reduce weights on less frequently utilized nodes. This dual mechanism allows our system to provide increasingly relevant recommendations while adapting to changing user interests. Experimental results demonstrate the effectiveness of our approach, showcasing improved recommendation performance compared to traditional methods.