Gender Bias in Spotify Recommendations

Presenter's Name(s)

Sydney White

Conference Year

2024

Abstract

As machine learning becomes more and more prevalent, so does the bias it breeds and perpetuates. Recommendation algorithms, in particular, govern much of our interactions with technology and content online, leading to significant consequences to various stakeholders if unjust. This project analyzes various types of gender bias in Spotify’s Discover Weekly recommendation algorithm by analyzing recommended output for experimental users of different genders and preferred genres. Ultimately, this experiment serves as an example of the kind of work required to assess recommender systems for bias and suggests a framework for doing so by considering stakeholders and their competing fairness objectives. Applying this framework on Spotify, the simulation suggests consistent bias on the basis of the producer’s gender acting on individual women and groups of women in Spotify recommendations as well as bias on the basis of the user’s gender. The effects of this bias differ by genre in users’ taste profiles.

Primary Faculty Mentor Name

Lisa Dion

Graduate Student Mentors

Jason Hibbeler

Status

Undergraduate

Student College

College of Arts and Sciences

Second Student College

College of Engineering and Mathematical Sciences

Program/Major

Data Science

Second Program/Major

Computer Science

Primary Research Category

Engineering and Math Science

Abstract only.

Share

COinS
 

Gender Bias in Spotify Recommendations

As machine learning becomes more and more prevalent, so does the bias it breeds and perpetuates. Recommendation algorithms, in particular, govern much of our interactions with technology and content online, leading to significant consequences to various stakeholders if unjust. This project analyzes various types of gender bias in Spotify’s Discover Weekly recommendation algorithm by analyzing recommended output for experimental users of different genders and preferred genres. Ultimately, this experiment serves as an example of the kind of work required to assess recommender systems for bias and suggests a framework for doing so by considering stakeholders and their competing fairness objectives. Applying this framework on Spotify, the simulation suggests consistent bias on the basis of the producer’s gender acting on individual women and groups of women in Spotify recommendations as well as bias on the basis of the user’s gender. The effects of this bias differ by genre in users’ taste profiles.