Development of a music recommendation system for motivating exercise

Abstract

While the health benefits of regular physical activity are well-established, many people exercise much less than is recommended by established guidelines. Music has been shown to have a motivational effect that can encourage people to exercise more strenuously or for longer periods of time, but the determination of which songs should be provided to which exercisers is an unsolved problem. We propose a system that incorporates user profiling to provide a strong set of initial recommendations to the user. Reinforcement learning is then used as each recommendation is accepted or rejected in order to ensure that subsequent recommendations are also likely to be approved. Test subjects who used the proposed system rated the playlists it provided more highly than those provided by a prior state-of-the-art reinforcement learning-based music recommendation system and also did not need to reject as many songs before being satisfied with their recommendations, both when receiving recommendations based on individual profiles, and when receiving recommendations based on aggregate profiles formed by grouping the users.