Algorithm-based music recommendations: Low accuracy for lovers of non-mainstream music

Algorithm-based music recommendations: Low accuracy for lovers of non-mainstream music


IMAGE

IMAGE: A examine revealed within the open-access journal ‘EPJ Knowledge Science’ means that music suggestions for high-energy music equivalent to onerous rock and hip-hop could also be much less correct than these for…
view more 

Credit score: Mockup: © TU Graz; Picture: © Lunghammer – TU Graz

A staff of researchers from Graz College of Know-how, Know-Middle GmbH, Johannes Kepler College Linz, College of Innsbruck, Austria and College of Utrecht, the Netherlands, in contrast how correct algorithm-generated music suggestions had been for mainstream and non-mainstream music listeners. They used a dataset containing the listening histories of 4,148 customers of the music streaming platform Final.fm who both listened to principally non-mainstream music or principally mainstream music (2,074 customers in every group). Based mostly on the artists music customers’ listened to most incessantly, the authors used a computational mannequin to foretell how seemingly music customers had been to love the music beneficial to them by 4 widespread music advice algorithms. They discovered that listeners of mainstream music appeared to obtain extra correct music suggestions than listeners of non-mainstream music.

Algorithm to categorise music listeners

The authors then used an algorithm to classify the non-mainstream music listeners of their pattern primarily based on the options of the music they most incessantly listened to. These teams had been: listeners of music genres containing solely acoustic devices equivalent to folks, listeners of high-energy music equivalent to onerous rock and hip-hop, listeners of music with acoustic devices and no vocals equivalent to ambient, and listeners of high-energy music with no vocals equivalent to electronica. The authors in contrast the listening histories of every group and recognized which customers had been the almost certainly to take heed to music exterior of their most popular genres and the variety of music genres listened to inside every group.

Those that principally listened to music equivalent to ambient had been discovered to be almost certainly to additionally take heed to music most popular by onerous rock, folks or electronica listeners. Those that principally listened to high-energy music had been least prone to additionally take heed to music most popular by folks, electronica or ambient listeners, however they listened to the widest number of genres, for instance onerous rock, punk, singer/songwriter and hip-hop,

The authors then used customers’ listening histories and a computational mannequin to foretell how seemingly the completely different teams of non-mainstream music listeners had been to love the music suggestions generated by the 4 widespread music advice algorithms. They discovered that those that listened to principally high-energy music appeared to obtain the least correct music suggestions and those that principally listened to music equivalent to ambient appeared to obtain probably the most correct suggestions.

Biased music advice algorithms

Elisabeth Lex, the corresponding creator, mentioned: “As growing quantities of music have turn out to be accessible through music streaming providers, music advice programs have turn out to be important to serving to customers search, kind and filter intensive music collections. Our findings recommend that many state-of-the-art music advice methods might not present high quality suggestions for non-mainstream music listeners. This might be as a result of music advice algorithms are biased in direction of extra widespread music, leading to non-mainstream music being much less prone to be beneficial by algorithms.”

“Additional,” added Elisabeth Lex, “our outcomes point out that the music preferences of those that principally take heed to music equivalent to ambient will be extra simply predicted by music advice algorithms than the preferences of those that take heed to music equivalent to onerous rock and hip-hop. Because of this they might obtain higher music suggestions.

The authors recommend that their findings might inform the creation of music advice programs that present extra correct suggestions to non-mainstream music listeners. Nevertheless, they warning that as their analyses are primarily based on a pattern of Final.fm customers their findings is probably not consultant of all Final.fm customers or customers of different music streaming platforms.

This analysis space is anchored within the Discipline of Experience “Data, Communication & Computing”, one among 5 strategic foci of TU Graz.

###

Disclaimer: AAAS and EurekAlert! usually are not liable for the accuracy of stories releases posted to EurekAlert! by contributing establishments or for using any data by the EurekAlert system.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *