Researchers in Austria and the Netherlands have discovered that new song recommendation algorithms (such as those used by Spotify or Last.fm) have found it difficult to find songs that hip-hop, hard rock or punk fans like. They pointed out that there may be deviations in the algorithms of these music style listeners.
The most commonly used applications for listening to music, such as Spotify, Last.fm Or own YouTube, With algorithms that can predict and show you new content Music you might like. In simple terms, it is a recommendation system through collaborative filtering: application They record the artists and genres that users listen to, and compare these results with like-minded audiences to understand other people’s preferences.
Therefore, it is correct to recommend the lovers of Lil Nas X to Post Malone, or, if you like Soleá Morente, they will recommend Rigoberta Bandini to you.
However, these algorithms are not perfect for subjective and humane things like artistic creation and musical taste. Therefore, a team of researchers from the Technical University of Graz, the Know-Center GmbH research center, the Johannes Kepler University of Linz, the University of Innsbruck (all from Austria) and the University of Utrecht (Netherlands) hope to proceed test. Suggested accuracy The audio generated by these algorithms is especially for listeners of music that is not very popular or well known to the public.The main results are published in the latest issue of the magazine EPJ Data Science, Are these algorithms They failed a lot More audience Hard Rock and Hip Hop, Not other music genres.
To verify this, the team used the historical records of the songs listened to by 4,148 users of the Last.fm platform, both from listeners who usually listen to more popular commercial music, and from listeners who prefer less-known artists ( 2,074 users per group).According to the artist that each user listens to most often, the research uses Calculation model Four different recommendation algorithms are used to predict whether they want a new song or a new singer. In this way, they confirmed that pop music listeners tend to receive more accurate and precise recommendations compared to less commercial audience groups.
After this, the author classifies non-commercial music listeners as Four groups, According to the characteristics of the music they listen to most.These groups are: listeners of music genres that only contain instruments Acoustic, Such as folk or singer creators; music very energetic Like punk or hip-hop music; very musical Hearing but no sound Like ambient music; and music Energetic but no sound Like electronic products. Therefore, the research is able to compare the historical records of each group and use computational models to determine which users are more likely to listen to music outside of their preferences and the diversity of music genres in each group.
Due to this classification, the study found that acoustic listeners without sound also tend to appreciate the other three groups (vibrant, dynamic songs without sound and original sound), and obtained more accurate suggestions through computational models. in the meantime, Vibrant music Are those who received Worst suggestion Although his team features the widest range of music genres-hard rock, punk, hardcore, hip hop and pop rock, it still employs various algorithms.
Popularity bias in the algorithm
Elisabeth LexThe co-author of this work and associate professor of applied informatics at the Technical University of Graz emphasized that music recommendation algorithms are already “essential” for users who want to search, select, and filter a collection of music applications.
Nonetheless, this suggests that the algorithm may not be able to provide recommendations for non-commercial music listeners. “This may be because these systems Preference for pop musicTo bring the artist out Mainstream He said: “They listen less.”
Finally, the authors suggest that their findings can be used as a basis for creating a music recommendation system to provide more accurate recommendations. However, they warned that their analysis was based on a sample of Last.fm users, which may not be representative for this platform or other music platforms.
Kovald Wait. “Supporting the Underground: Beyond the Characteristics of Mainstream Music Audiences”. EPJ Data Science (2021). DOI: 10.1140 / epjds / s13688-021-00268-9