A group of sound and computing experts performed a study of music streaming apps, which revealed that results generated by algorithms were not always fair. The study which was conducted in 2020, was presented in March 2021 at the Conference on Human Information Interaction and Retrieval held in Canberra ACT Australia. It presented evidence that most digital, AI supported music recommenders are partial to male artists.
According to study authors Xavier Serra, Andres Ferraro and Christine Bauer, 400 million people subscribed to one music streaming app in 2020. The technology used a recommendation algorithm that is inclined to pick and recommend music of male artists over female female artists. They said that such partiality also exists in music streaming services, where only a handful of female superstars have taken prominence among the most popular artists in the music industry.
Although the researchers acknowledge that the problem takes roots from beyond the industry, the algorithms of online music platforms known as recommenders can also have an influential role. Their testing of the algorithms revealed that on the average, the first six recommended tracks are those of male artists. Songs of female musicians are queued by the 7th or 8th song.
In conjunction, the research group analyzed the listening behaviour of about 330,000 users for the past nine years, from which they were able to extract data that users had listened to only 25% of female artists rated as popular.
About the Study Authors
The sound and computing experts who conducted the research study include Professor Xavier Serra of Universitat Pompeu Fabra (UPF) Barcelona, who delves into research in the field of Sound and Music Computing. He is also the founder of UPF’s Music Technology Group.
Andres Ferraro is a UPF PhD student and doing research work at the university’s Department of Information and Communication Technologies Audio Signal Processing Lab.
Christine Bauer Christine Bauer is an assistant professor at the Human-Centred Computing group at the Department of Information and Computing Sciences at Netherland’s Utrecht University