A team of researchers at Know-Center, Graz University of Technology, Johannes Kepler University Linz, University of Innsbruck and University of Utrecht (the Netherlands) analysed the accuracy of algorithm-generated music recommendations for listeners of mainstream and non-mainstream music. They used a dataset containing the listening histories of 4,148 users of the music streaming platform Last.fm who listened mostly to non-mainstream music or mostly mainstream music.
„Since more and more music is becoming available via music streaming services, music recommendation systems have become essential for helping users to search, sort and filter extensive music collections. However, the quality of many cutting-edge music recommendation techniques for non-mainstream music users still leaves a lot to be desired. In our study we established that the willingness of a music user to listen to the kinds of music outside of his or her primary music preferences has a positive effect on the quality of recommendations. Thinking out of the box pays off even when it comes to listening to music,“ explained Dominik Kowald, first author of the study and Head of Research Area Social Computing at Know-Center.
Algorithm categorizes music listeners
“First, we established a computational model for artists, to whom music users listened most frequently. Based on this model, we were able to predict the probability of music users’ liking the music recommended to them by four common music recommendation algorithms. We found that listeners of mainstream music appeared to receive more accurate music recommendations than listeners of non-mainstream music“, stated Elisabeth Lex, the study’s scientific leader and Associate Professor of Applied Computer Science at Graz University of Technology.
Non-mainstream music listeners have been assigned by an algorithm into the following categories based on the types of music they most frequently listened to: users of music genres with only acoustic instruments (e.g. folk), users of high-energy music (e.g. hard rock, hip-hop), users of music with acoustic instruments and no vocals (e.g. ambient), and users of high-energy music with no vocals (electronica).
By comparing each group’s listening histories, the researchers identified users who were most likely to listen to music outside of their preferred genres and the diversity of music genres listened to by each group. Those who mainly listened to music such as ambient were found to be most likely to listen to music preferred by hard rock, folk and electronica listeners as well. Those who mainly listened to high-energy music were least likely to listen to music preferred by folk, electronica and ambient listeners. However, they listened to the widest variety of genres, e.g. hard rock, punk, singer/songwriter and hip-hop.
Biased music recommendation algorithms
By using the computational model, the researchers predicted the probability of various groups of non-mainstream music listeners’ liking the music recommendations generated by the four common music recommendation algorithms. They found that those who listened mainly to high-energy music appeared to receive the least accurate music recommendations and those who mainly listened to ambient get the most.
The authors pointed out potential benefits of their findings for the development of music recommendation systems that will provide more accurate recommendations to non-mainstream music listeners. However, they caution that since the study was based on a sample from Last.fm users, the results may not be representative of all Last.fm users and users of other music streaming platforms.
This research is anchored in the Field of Expertise Information, Communication & Computing, one of the five Fields of Expertise at TU Graz.
