Mining for Context in Playlist Names to Improve Music Recommendations

Goal of this thesis is to mine for information about the listening context in playlist names of playlist created by Spotify users. In a first step, a filter for non-contextual information, i.e., track- artist- or album names facilitating the Spotify API is implemented and applied to the data. In a second step, Google's pre-trained word2vec model is leveraged to estimate similarities between the vector representation of the playlists. Moreover, different clustering algorithms to find contextual similar playlists are applied to the vector representations. Finally, qualitative and quantitative analyses should give valuable insights for improving context-aware music recommender systems, i.e., for which events, occasions or activities users create playlists.