Comparative Analysis of Content-Based Recommender Systems Using Distance Metrics and Feature Sets for Classical Music
Date of Award
Master of Science in Chemistry, Straight Program
Information Systems & Computer Science
Andrei D. Coronel, PhD
Music recommender systems have become a popular tool utilized by numerous online music streaming apps like Spotify and Apple Music. De- spite the prevalence of music recommenders, not many have created one particularly for classical music. Although listeners of classical music are not typically dominant, they still constitute as a significant target group for music recommender systems. Classical music will greatly benefit from the use of a content-based recommendation system that will analyze the music’s rhythmic, melodical, and chordal features as these features help define a user’s musical taste. As such, we present an approach for content-based recommendation using similarity of classical music using high-level musical features. The chosen high-level musical features include rhythmic variability, chromatic motion, melodic embellishments, and key. Comparison of vari- ous feature selection and processing is used and experimented on to min- imize the computational expense of the method while maximizing results. Finally, the paper compares different evaluations metrics that represent the effectiveness of the recommendations through a listening test. The results demonstrate the feasibility of these features and techniques in creating a content-based recommender for classical music.
(2019). Comparative Analysis of Content-Based Recommender Systems Using Distance Metrics and Feature Sets for Classical Music. Ateneo de Manila University.