Analyzing the Characteristics of Music Playlists using Song Lyrics and Content-based Features

Thesis Type Master
Thesis Status
Student Stefan Wurzinger
Thesis Supervisor
Research Field

Today's online music platforms offer you the ability to listen to personalized playlists or radio stations. The collaborative filtering (CF) technique is widely used to offer such services. A well-known disadvantage of CF is that new and unpopular songs can't be recommended (cold-start problem). To mitigate the cold-start problem, song lyrics and content-based features can be considered when recommending music. Therefore, the goal of this thesis is to analyze the characteristics of music playlists to find some significant lyrics and content-based features which can be used to decide if a song belongs to a specific playlist or not. Furthermore, it should be investigated if the extracted features can be used for music recommendation or automatic playlist generation.