Adventurous Genre Exploration

Music recommender systems based on collaborative filtering tend to be poorly suited for discovering music in the long tail and are ill suited for exploring new or unrelated genres. Content-based filtering has been shown to work better for recommending similar music in the long tail. This thesis proposes and evaluates a novel extension of content-based filtering that takes into account how adventurous a user is when it comes to exploring new music genres. To this end, an extensible recommender system and web-based user interface are built on top of the Spotify API. A user study is then conducted in order to investigate to which extent a user model based on the most listened tracks on Spotify combined with this recommender provides an effective way of discovering likable new music in genres that the user otherwise doesn’t listen to.