Summary: Commonly used music recommendation algorithms tend to favor male artists, reducing visibility and exposure for female musicians. A recent study from researchers at UPF Barcelona and Utrecht University examines this gender imbalance and proposes a re-ranking method to mitigate the bias.
Source: UPF Barcelona
Music industry inequalities exist offline—and recommendation systems online can amplify them.
Researchers Andrés Ferraro and Xavier Serra from the Music Technology research group (MTG) at the UPF Department of Information and Communication Technologies (DTIC), together with Christine Bauer from Utrecht University, investigated how popular music recommendation algorithms handle artist gender and whether they worsen existing disparities. Their work responds to concerns raised directly by artists about fair access to exposure and the role that platform algorithms play in shaping listening habits.
The study began with interviews of music creators. Those conversations highlighted that gender fairness is a key concern for artists: many feel female performers receive less visibility and fewer opportunities through algorithmic recommendation than their male counterparts.
Women receive less exposure in algorithmic rankings
To quantify these observations, the team evaluated a widely used collaborative filtering recommender on two public datasets enhanced with gender labels. They found that the recommender tends to reproduce the gender imbalance present in the data: when only a minority of artists are female, recommended rankings mirror that skew and place women lower in recommendation lists.
On average, the first recommended female artist appeared several positions later in the ranked list than the first recommended male artist—indicating that female artists are less likely to appear near the top of algorithmically generated suggestions. This lower placement translates directly into reduced exposure, since users frequently listen to the top items in a list.
Ferraro, the paper’s first author, explains that this exposure gap stems from how recommendations are generated: collaborative methods learn from historic listening patterns and then prioritize items that match those patterns. When the training data contains fewer listens to female artists, the model naturally ranks male artists higher, producing a feedback loop as recommendations shape user listening and reinforce the original imbalance.
The feedback loop is particularly concerning because it amplifies the initial bias: as the system recommends mostly male artists, user behavior adapts to those recommendations, providing further biased input to the algorithm and making the imbalance harder to correct over time.
Re-ranking as a practical way to improve gender balance
To address this problem, the authors propose a progressive re-ranking strategy that adjusts recommendation lists to increase the visibility of underrepresented groups. Instead of changing the underlying collaborative filtering model, the method reorders the top results to ensure a higher presence of female artists while preserving relevance. In simulation experiments that model feedback loops, this re-ranking led to measurable changes in simulated user behavior: over time, users exposed to re-ranked lists listened to a greater share of female artists than users receiving conventional recommendations.
The approach is designed to be practical: it can be applied on top of existing recommendation pipelines and does not require altering user data collection. By systematically promoting a set number of positions for female artists within recommendations, the system counteracts the tendency of collaborative filtering to reproduce historical imbalances and helps establish a new, fairer listening pattern.
About this AI and gender discrimination research news
Source: UPF Barcelona
Contact: Núria Pérez – UPF Barcelona
Image: The image is in the public domain
Original Research: Open access. “Break the Loop: Gender Imbalance in Music Recommenders” by Andrés Ferraro, Xavier Serra, Christine Bauer. Proceedings of the 2021 Conference on Human Information Interaction and Retrieval
Abstract
Break the Loop: Gender Imbalance in Music Recommenders
As recommender systems play an important role in everyday life, there is growing pressure for these systems to act fairly. Beyond serving users with diverse tastes, recommenders must also represent content providers—such as musicians—equitably. Artists who contributed to this study emphasized that female creators should receive more exposure through music recommendations.
The authors analyze a commonly used collaborative filtering approach on two public datasets that include gender information, evaluating how the method performs with respect to artist gender. To improve gender balance, the paper proposes a progressive re-ranking method inspired by artist interviews. Evaluation uses a simulation of feedback loops and a detailed set of performance and fairness metrics to assess long-term effects.
The findings underline that algorithmic choices matter: interventions at the ranking stage can help counteract historical biases and produce fairer exposure for underrepresented artists without sacrificing recommendation quality. This work highlights a practical path forward for platforms seeking to reduce gender imbalance and promote a more diverse musical ecosystem.