Large Language Models for Recommendation: Progresses and Future Directions

Abstract

Large language models (LLMs) have significantly influenced recommender systems. Both academia and industry have shown growing interest in developing LLMs for recommendation purposes, an approach commonly referred to as LLM4Rec. This involves efforts such as utilizing LLMs for generative item retrieval and ranking, along with the potential for creating universal LLMs for varied recommendation tasks, signaling a possible paradigm shift in recommender systems. This tutorial is designed to review the progression of LLM4Rec and provide an in-depth analysis of the prevailing studies. We will discuss how LLMs advance recommender systems in model architecture, learning paradigms, and capabilities like conversation, generalization, planning, and content generation. Additionally, the tutorial will highlight open problems and challenges in this nascent field, addressing concerns related to trustworthiness, efficiency, online training, and recommendation data modeling. Concluding with a summary of the takeaways from previous research, the tutorial will suggest avenues for future investigations. Our aim is to help the audience grasp the developments in LLM4Rec, as well as to spark inspiration for further research. By doing so, we expect to contribute to the growth and success of LLM4Rec, possibly leading to a fundamental change in recommender paradigms.