Conversational recommender systems with large language models: A review of current developments

초록

Conversational Recommender System (CRS) aims to understand users’ real-time preferences through natural language interactions and provide personalized recommendations with explanations. The recent advent of Large Language Models (LLMs) has shown promising potential for effectively capturing users’ subtle and implicit preferences through human-like fluent interactions, leading to rapid progress in CRS research. Notably, the application of LLMs is expanding beyond the singular role of a recommendation engine, with multifaceted integrations across various components of the CRS architecture-such as data augmentation, system control, and evaluation. To systematize this diverse and rapidly evolving landscape, this survey provides the first comprehensive review of LLM-based CRS, categorizing research according to their utilization purposes and integration approaches. We classify and analyze three key paradigms: (1) LLM as Conversational Recommender, which directly utilizes LLMs as core recommendation engines; (2) LLM-Augmented CRS, which leverages LLMs to enhance specific system components and processes; (3) LLM-Assisted CRS Evaluation, which utilizes LLMs for user simulation and CRS performance assessment. We provide in-depth analysis by further subdividing each paradigm into detailed categories, discussing their key approaches and advancements. Furthermore, we analyze technical and ethical challenges and outline promising research directions for advancing LLM-based CRS development.

키워드