This project is part of my Bachelor's thesis for my B.Sc. Internet Computing at the University of Passau.
Thanks to the rise of music streaming services, such as Spotify, Apple Music, or Pandora, anybody can instantly access almost all the music in the world. This has led to increased demand in using recommendations to help users find relevant and interesting music tracks and playlists. Research on recommender systems (RSs) has already gained interest throughout the last decades and resulted in some valuable techniques for general recommendations. However, the field music recommender systems (MRSs) in particular is still only sparsely researched. One emerging approach is incorporating evolutionary computation (EC) and swarm intelligence (SI) techniques in the recommendation process. Notwithstanding, using these biology-inspired approaches explicitly in MRSs lacks research as well.
This thesis aims at partially filling this gap in research by examining some EC and SI techniques in the field MRSs. For that, I lay out some key learnings from existing research on EC and SI in general RSs, analyse possible application areas in MRSs, and compare different algorithms for the MRS task of music playlist continuation (MPC). This task has gained much attention during the RecSys Challenge 2018 and was identified as one of the grand challenges in MRSs. The primary focus of this thesis lies in designing, implementing, and conducting a study that compares EC and SI algorithms in the context of the MPC task. The study results show an overall inferior solution quality of the examined algorithms compared to machine learning approaches. Nevertheless, the EC algorithm ”SMS-EMOA” yields the best results among the other algorithms in the study.
The study results suggest that using EC and SI algorithms for calculating and optimising solutions in the MPC task is not very suitable. However, using EC and SI in other (music) recommendation application areas highlighted in this thesis still seems promising for future work.