Improving Fairness and Transparency for Artists in Music Recommender Systems

Abstract

Streaming services have become one of today's main sources of music consumption, with music recommender systems (MRS) as important components. The MRS' choices strongly influence what users consume, and vice versa. Therefore, there is a growing interest in ensuring the fairness of these choices for all stakeholders involved. Firstly, for users, unfairness might result in some users receiving lower-quality recommendations in terms of accuracy and coverage. Secondly, item provider (i.e. artist) unfairness might result in some artists receiving less exposure, and therefore less revenue. However, it is challenging to improve fairness without a decrease in, for instance, overall recommendation quality or user satisfaction. Additional complications arise when balancing possibly domain-specific objectives for multiple stakeholders at once. While fairness research exists from both the user and artist perspective in the music domain, there is a lack of research directly consulting artists - with Ferraro et al. (2021) as an exception. When interacting with recommendation systems and evaluating their fairness, the many factors influencing recommendation system decisions can cause another difficulty: lack of transparency. Artists indicate they would appreciate more transparency in MRS - both towards the user and themselves. While e.g. Millecamp et al. (2019) use explanations to increase transparency for MRS users, to the best of our knowledge, no research has addressed improving transparency for artists this way.