Music Playlist Generation for Emotion Regulation

Abstract

Dementia care is a growing problem, both due to a rising number of cases and due to a shortage in healthcare workers. Aside from cognitive symptoms persons with dementia (PwDs) often deal with psychological symptoms such as agitation. The individualized music intervention (IMI) by Linda Gerdner has been proposed to reduce these. This is the basis for our first research question “How can we incorporate the IMI into technology to care for PwDs?". We answer this by dividing the guidelines for IMI into functional components with their own responsibilities. This allows for more independent development of these components. We also research one of the components, a playlist generator. This leads us to our second research question "Assuming knowledge of a list of preferred music for a user, can we generate playlists which better regulate emotions than a random shuffle?". Our generator is based on a Markov Decision Process (MDP) and allows for a target to be set in the 2D valence-arousal space. This was tested in an online experiment with 22 participants. The control group listened to a random shuffle of their preferred music. The experimental group listened to a playlist generated by the MDP. The target was set to a valence > 0.75 (valence ranging from 0 to 1). A significantly higher final valence for the experimental group was observed, showing a clear improvement over the control group when the goal is to positively influence the valence. The dataset obtained through this experiment is available for future research.