How can recommendation algorithms produce new inequalities or reinforce existing ones?
According to the Algo_Ritimos project, recommendation algorithms tend to reinforce gender inequality in the Brazilian music scene. Additionally, streaming apps do not provide data on the ethnic and racial identities of artists and users.
Today (8th), InternetLab attended RightsCon, the largest human rights event in the digital era, which this year took place in San Jose, Costa Rica, to talk about Algo_Ritmos, a project developed with support from the Equity Fund of the Wikimedia Foundation’s Knowledge Equity Program. The project aims to understand how algorithmic recommendations produced by music streaming platforms such as Spotify and Deezer influence and are influenced by structural inequalities.
The past decade has witnessed the emergence and global growth of streaming platforms, such as Spotify and Deezer, in the music industry. These platforms have transformed cultural consumption, taste formation, and the way people explore and discover new artists. Algo_Ritmos is the result of an investigation conducted to identify any biases that may be reinforced or produced in these contexts, as well as to reflect on the impacts of the globalization of these platforms and, consequently, the normalization of cultural uses and consumption mediated by this type of technology.
With that, we selected the Spotify and Deezer platforms to begin this investigation. In order to do so, we sought to understand how algorithmic recommendations produced by both platforms can operationalize structural inequalities in their recommendations. Thus, from an intersectional perspective, we analyzed the data collected from these platforms to understand how gender, race, and ethnicity dimensions are present in the recommendation of music and artists.
As we could not find data related to the ethnic and racial background of artists, the Algo_Ritmos website shows the main differences observed when looking at music consumption in Brazil and Brazilian artists. It is also possible to simulate the results mentioned above in practice. The publication is available in Portuguese and English.