A content-based music recommender system

Abstract

Music recommenders have become increasingly relevant due to increased accessibility provided by various music streaming services. Some of these streaming services, such as Spotify, include a recommender system of their own. Despite many advances in recommendation techniques, recommender systems still often do not provide accurate recommendations.

This thesis provides an overview of the history and developments of music information retrieval from a more content-based perspective. Furthermore, this thesis describes recommendation as a problem and the methods used for music recommendation with special focus on content-based recommendation by providing detailed descriptions on the audio content features and content-based similarity measures used in content-based music recommender systems. Some of the presented features are used in our own content-based music recommender.

Both objective and subjective evaluation of the implemented recommender system further confirm the findings of many researchers that music recommendation based solely on audio content does not provide very accurate recommendations.