Algorithmic Assortative Matching on a Digital Social Medium

Online algorithms recommend “people we may know” and “content we may like.” Inherent in these recommendations is a notion of positive assortativity in which the people and content being suggested to us match our own preferences and beliefs. In this paper, we focus on such tacit (i.e., behind the scenes) algorithmic facilitation of assortativity at work across digital platforms and social media. To investigate the effects that it has on human online relating and behavior, we conduct a large-scale field experiment in a mobile social game in which we switch algorithmic assortative matching between new users and existing communities on and off over the course of six weeks. With the help of model-based analysis, we find such assortative matching to increase firm profits (measured as user engagement and monetization) via increased sociality (measured as user messaging). Results further show that such behind-the-scenes algorithmic matching leads to a segregating path between engaged and marginal online communities, further marginalizing less engaged and connected users. Our findings, hence, pinpoint a conflict between profit-centered and societally equitable management of online platforms and are important toward more algorithmic transparency and fairness as online algorithms structure ever larger parts of human life. Humans are increasingly interacting in and operating their daily lives through structured digital and virtual environments, mainly through apps that provide media for sharing photos, messaging, gaming, collaborating, or video watching. Most of these digital environments are offered under “freemium” pricing to facilitate adoption and network effects. In these settings, users’ early social interaction and experience often have a substantial impact on their longer term behavior. On this background, we study the impact of an algorithmic system that matches new users to existing communities in an assortative manner. We devise a machine learning–based matching system that identifies users with high expected value and provides them the option to join highly active, in terms of engagement and expenditure, teams. We deploy this mechanism experimentally in a digital social game and find that it significantly increases user engagement, spending, and socialization. This finding holds for more active communities and overall. Teams matched with low-activity new users are negatively impacted, leading to an overall more segregated social environment. We argue that