Social Identity Theory and Algorithmic Bias: Ingroup and Outgroup Acrophily in Recommender Systems

Recommender systems exert a substantial influence on the exposure to political content; however, their role in social polarization remains inadequately explored. This study investigates whether these systems exacerbate biases derived from Social Identity Theory: ingroup acrophily (the preference for extreme content that aligns with one’s political views) and outgroup acrophily (the exposure to extreme opposing content). By analyzing over 300,000 YouTube videos and 1.7 million recommendation links, we construct a recommendation network and employ permutation-based null models alongside Exponential Random Graph Models to identify deviations from anticipated patterns. Our findings provide compelling evidence of ingroup and outgroup acrophily: Center-Right users receive a greater volume of extreme right-wing recommendations (ingroup acrophily), while users occupying both ideological extremes encounter extreme content from the opposing side (outgroup acrophily). Emotional analyses indicate that ingroup acrophily arises from feelings of anger and uncertainty regarding institutional and moral issues, while outgroup acrophily is motivated by emotions such as disgust and sadness, particularly concerning culture and identity. The study presents four key insights: it introduces the concept of outgroup acrophily, demonstrates that algorithmic biases are amplified by platform design rather than simply mirroring user preferences, reconceptualizes acrophily as a sociotechnical phenomenon heightened by algorithms, and applies Social Identity Theory to the digital context, thereby elucidating how these systems influence intergroup dynamics and contribute to political polarization. Practically, our findings inform the design of diversity-aware and fairness-sensitive algorithmic solutions, with recommender systems serving as a critical application domain. These insights help both platform designers and policymakers develop algorithmic infrastructures that foster diversity and minimize polarization.