Modeling User Preferences for Music Recommendations with LLMs: Opportunities and Limitations

Room A008

This talk explores the potential of large language models (LLMs) in enhancing music recommendation systems through explicit modeling of user preferences. We begin by comparing traditional recommenders with the capabilities introduced by LLMs. The discussion will focus on strategies for capturing preferences within conversations and extending to long-term user modeling beyond conversations. While we will examine the effectiveness of LLMs in various tasks, such as query parsing and user preference summarization, we will also address their limitations. The goal is to provide a comprehensive view of the challenges involved in modeling user preferences from and through text in the music domain. Bios: Elena Epure is a Senior NLP Research Scientist at Deezer Research. Elena’s current research focuses on creative media, building on her previous work with social media and news content. Her work integrates NLP fundamentals, including semantics, text mining, and information extraction, to advance creative media understanding, personalization, and contextualization. She also explores topics relayed to conversational and natural language recommendation and search. Bruno Sguerra is a research scientist at Deezer Research, specializing in understanding user behavior in music streaming. His work spans search intent analysis, modeling listening contexts, and enhancing music recommendation systems. Currently, he is exploring the music discovery process and developing methods to denoise user feedback, aiming to improve the accuracy and relevance of recommendations.