DalSegno: User-centric preference elicitation strategies for mitigating cold start in music recommender systems

Abstract

Avid music enthusiasts often rely on music recommender systems to sift through expansive music catalogs and find new songs fitting their interests. However, such systems struggle to personalize suggestions for new users as they heavily rely on extensive listening histories to make accurate suggestions — an issue known as the new user cold start problem. This problem is exacerbated by the fact that most commercial recommender systems lack transparency and avenues for users to influence their recommendations.

We thus propose DalSegno, a music recommender system with an interactive web-based user interface. The platform is designed to overcome the new user cold start problem by iteratively presenting users with recommendations and incorporating elicited feedback. Additionally, DalSegno enables users to learn about and fine-tune their inferred music preferences through interactive visualizations of song characteristics.

Throughout three rounds of user testing, DalSegno demonstrated promising results. Participants appreciated the system's ability to incorporate user feedback to provide more relevant recommendations and considered it more intuitive to use than commercial recommendation systems. Additionally, users felt that the interactive visualizations of musical qualities helped them learn more about their personal music tastes, which encouraged them to further utilize the interface. Overall, positive evaluations of DalSegno demonstrate that incorporating user input and fostering explainability is vital to creating a more user-focused and effective music discovery experience.