Fusion-based Music Recommender System Using Music Affective Space based on Serendipity

Abstract

In recent years, as the popularity of music distribution services has increased, research on music recommendation has become more active. In particular, factors not only accuracies but also contextual information such as user’ $s$ emotion and “serendipity” are also considered necessary to improve the quality of music recommendations and attract attention. Serendipity is defined as novelty, unexpectedness, and preference; these items are considered important factors as evaluation criteria for recommender systems. In this paper, we propose a system that recommends music, named FUSION MUSIC, based on the affective information of two favorite music the user selects. The FUSION MUSIC is based on two concepts; one is a music affective space that reflects the affective information of each music, and another is a fusion-based music recommendation method that creates a partial affective space within that space and recommends serendipitous music. We finally developed FUSION MUSIC using these concepts and the Spotify API. The results of the evaluation experiment showed that FUSION MUSIC has the potential to recommend more serendipitous music compared to Spotify. For the proposed system, we will investigate more detailed validation methods in the future.