Psychology-Inspired Machine Learning for Music: Evoked Emotion Recognition and Emotion-Aware Music Recommendation

Music, and the way it is composed, listened to, and even advertised and sold, has always been tightly connected to the psychological states it evokes. Already in 1921, the early ages of record music marketing [Pelly 2025], Thomas A. Edison collected data regarding the changes in the emotional states of music listeners. The data served as a basis to create music collections aimed at influencing the listeners' emotional states, and meant to be sold as records. ¹ Today, more than one hundred years later, large-scale music streaming platforms offer mood playlists that substantially contribute to the total number of streams. The tendency to use music to regulate one's emotional state is not only evident from the commercial choices of music distributors, it has also been backed up by scientific research [Boer and Fischer 2011; Saarikallio et al. 2013; Schäfer et al. 2013]. Therefore, the ability of music to evoke emotions has been a topic of active research in the field of psychology of music [Juslin and Laukka 2004; Zentner et al. 2008; Vigl and Zentner 2023; Jacobsen et al. 2025].