Modeling Diversity of Music Recommendation via Multimodal Features

Music recommender systems allow us to find music we like from the large collections of music available via streaming services such as Spotify. To this end, research has mostly focused on recommending the most relevant items for users. However, factors like diversity and novelty of recommendation lists also play an important role for users. In this thesis, we aim to investigate the diversity of music recommendation lists and advance current diversification algorithms based on multimodal features (audio, embedded metadata, expert-generated content, user-generated content, etc.). Importantly, this data allows optimizing the diversity of a recommendation list along multiple dimensions.