These days, more and more people listen to music on streaming apps – in early 2020, 400 million people were subscribed to one. These platforms use algorithms to recommend music based on listening habits. The recommended songs might feature in new playlists or they might start to play automatically when another playlist has ended.
But what the algorithms recommend is not always fair. In a new study we showed a widely used recommendation algorithm is more likely to pick music by male than female artists. In response, we've come up with a simple way to give more exposure to female artists.
The representation of women and gender minorities in the music industry is tremendously low. About 23% of artists in the 2019 Billboard 100 were women or gender minorities. Women represent 20% or less of registered composers and songwriters, while 98% of works performed by major orchestras are by male composers.
This bias is also present in streaming services. A few female "superstars" dominate among the most popular artists, but most female and mixed-gender artists are in the lower levels of popularity. While the problem stems from beyond the music industry, online music platforms and their algorithms that recommend music – called recommenders – play a large role.
While previous studies have repeatedly asked consumers for their opinion, the music artists, those providing the content, are rarely in the loop.
We wanted to put the spotlight on artists. We asked musicians to give us their views on what would make online music platforms more fair. When they said gender imbalance was a major problem, we decided to study this in more detail.
Our analysis of around 330,000 users' listening behaviour over nine years showed a clear picture – only 25% of the artists ever listened to were female. When we tested the algorithm we found, on average, the first recommended track was by a man, along with the next six. Users had to wait until song seven or eight to hear one by a woman.
Breaking the loop
As users listen to the recommended songs, the algorithm learns from these. This creates a feedback loop.
To break this feedback loop, we came up with a simple approach to gradually give more exposure to female artists. We took the recommendations computed by the basic algorithm and re-ranked them – moving male artists a specified number of positions downwards.
In a simulation, we studied how our re-ranked recommendations could affect users' listening behaviour in the longer term. With the help of our re-ranked algorithm, users would start changing their behaviour. They would listen to more female artists than before.
Eventually, the recommender started to learn from this change in behaviour. It began to place females higher up in the recommended list, even before our re-ranking. In other words, we broke the feedback loop.
This shows how easy it can be. Our simple method can help address the biases in the algorithms that play a large role in the way many people discover new music and artists. Next, we hope to study how real consumers perceive the changes introduced by the re-ranking strategy and how it impacts their listening behaviour in the long term.
Another crucial step would be to collect and use data about the wide scale of gender identities. We're aware this binary gender classification does not reflect the multitude of gender identities. The unavailability of data beyond the gender binary is a massive obstacle, both for research as well as for taking action and making progress on a societal level.
So far, our simulation could demonstrate the benefits of a simple re-ranking approach. But responsibility is, of course, not with the platform providers alone. Initiatives such as Keychange and Women in Music are working to represent the underrepresented in the music industry. The rest of us need to follow.
As music festivals are being criticised for the lack of women in their lineups, any step towards representing more women all genders in a more balanced manner is a step in the right direction.
This article is republished from The Conversation under a Creative Commons license.