Come Together: How Social Agents can Improve Music Discovery

Author

Laura Triglia, Pietro Gravino, Thomas Carette, Francesco Rea, Alessandra Sciutti, and Pablo Barros

Abstract

Traditional recommendation interfaces often struggle to motivate users to explore beyond their established preferences. This study examines whether a conversational, LLM-powered social agent can encourage engagement with unfamiliar music during playlist cocreation. The agent first builds rapport over known preferences, then suggests both familiar and unfamiliar songs. In a user study, participants interacted with both a chat-based agent and a traditional form interface, selecting songs “close to” or “far from” their usual tastes. We analyzed song choices, user experience, and perceived social dynamics. While the agent did not significantly increase the selection of unfamiliar songs, it enhanced enjoyment of both familiar and unfamiliar tracks. Linguistic analysis indicates that longer engagement with the agent correlates with greater appreciation of novel recommendations. Despite some usability challenges, users found the agent more engaging than forms. These findings suggest socially aware recommendation systems can improve user experience and foster music exploration.