Algorithmic Impact Assessments at Scale: Practitioners’ Challenges and Needs
Amar Ashar, Karim Ginena, Maria Cipollone, Renata Barreto, Henriette Cramer
This study examines gender representation in current music streaming, utilizing one of the world’s largest streaming services. First, we found listeners generally stream fewer female or mixed-gender creator groups than male artists, with differences per genre. Second, while still relatively low, we found that recommendation-based streaming has a slightly higher proportion of female creators than “organic” listening (ie, tracks that are not recommended by editors or algorithms). Third, we examined streaming data from 200,000 US users to determine the proportion of female artists in organic and recommended streams over a 28-day period and the relationship between recommended streams and users’ future organic listening. The proportion of female artists in recommended streaming appears predictive of the proportion of female artists in organic streaming; these effects are moderated by gender and age. Fourth, this study also samples creators across different popularity levels, seeing more female and multi-gender groups at lower levels than in the middle tiers. However,(solo) female artists are better represented again in the superstars category, suggesting influence of selected superstars and genres. We conclude by discussing potential avenues in algorithmic
Amar Ashar, Karim Ginena, Maria Cipollone, Renata Barreto, Henriette Cramer
Yijun Tian, Maryam Aziz, Alice Wang, Enrico Palumbo and Hugues Bouchard