Algorithmic Glass Ceiling: The effect of social recommender systems on diversity

Social recommendations (friend suggestion, people to follow, and the like) were shown to affect the network growth of social media. Simultaneously, a growing concern has documented signs of intrinsic barriers to equal opportunity online, either due to decisions informed by algorithms using personal data, or even in the spontaneous growth of interactions that online services facilitate. Leveraging new data collected from Instagram, we offer here for the first time an analysis that studies the effect of gender, homophily and growth dynamics completed with the effect of social recommendation algorithms. Our main finding is that prominent social recommendation algorithms, under natural conditions, emph{exacerbates} the under-representation of demographic groups at the top. We prove, empirically and through mathematical analysis, the presence of an emph{algorithmic glass ceiling}, exhibiting all properties of the metaphorical barrier preventing subgroups to reach superior notoriety. What raises largest concerns is that we mathematically prove, under fixed minority and homophily parameters, that the algorithmic effect is systematically larger than the glass ceiling generated by the spontaneous growth of social networks. We briefly discuss ways to explore to address this concern in future design.