Systems musical recommendation that run on algorithms, such as those on streaming music platforms, could make less accurate suggestions for certain users, depending on the music they listen to.
This conclusion was reached by a recently published investigation, which analyzed a sample of listening data obtained from Last.fm.
The difference was marked between the suggestions that listeners of music of styles such as hard rock or hip-hop can receive, which turned out to be less precise compared to the recommendations received by those people who are more assiduous to less conventional musical variants , which turned out to be more successful.
The research was developed by a team from the Graz University of Technology, Know-Center GmbH, the Johannes Kepler University of Linz, the University of Innsbruck, Austria and the University of Utrecht, the Netherlands.
This analysis carried out on the work of the algorithms at the time of making recommendations, used for its research a data set composed of listening histories of 4148 users of the Last.fm platform. This sample was divided into two equal groups, of 2074 users each, separated by the predominance of “conventional” or “unconventional” music streams.
From that set of data, the artists listened to most frequently by users were filtered, to submit them to a computational analysis model, in order to predict how likely it is that the recommendations will end up being liked by their recipients.
The task of classifying listeners under a criterion of predominance of “unconventional” musical preferences was developed with the help of an algorithm. These groups are: listeners of music genres that contain only acoustic instruments such as folk songs, listeners of high-energy music (such as the aforementioned hard rock and hip-hop), listeners of music with acoustic instruments without a human voice, and electronic music. The counterpart of this group was formed based on the remaining musical genres.
Subsequently, the playback histories of each group were compared to determine which users like to listen to music outside of their favorite artists and the variety of musical genres that are heard in each group.
The results of the analysis revealed that the algorithmic recommendations were found to be more effective in the group of listeners of “unconventional” music.
Research team observations
Elisabeth Lex, lead author of the study, commented that these results “They indicate that the musical preferences of those who primarily listen to music such as ambient can be more easily predicted by music recommendation algorithms than the preferences of those who listen to music such as hard rock and hip-hop”.
More in depth, Lex added that “As increasing amounts of music have become available through music streaming services, music recommendation systems have become essential to help users search, sort and filter large music collections. State-of-the-art music recommendation techniques may not provide quality recommendations for unconventional music listeners. This could be because the music recommendation algorithms are biased towards popular music, making the algorithms less likely to recommend unconventional music ».
An important precision, discussed in the study, is that the findings made may not be fully representative, basing their results solely on data from only one source.
The publication of this study is available in the journal EPJ Data Science.