A high-dimensional approach to measuring online polarization

Abstract

Abstract Polarization, ideological and psychological distancing between groups, can cause dire societal fragmentation. Of chief concern is the role of social media in enhancing polarization through mechanisms like facilitating selective exposure to information. Researchers using user-generated content to measure polarization typically focus on direct communication, suggesting echo chamber-like communities indicate the most polarization. However, this operationalization does not account for other dimensions of intergroup conflict that have been associated with polarization. We address this limitation by introducing a high-dimensional network framework to evaluate polarization based on three dimensions: social, knowledge, and knowledge source. Following an extensive review of the psychological and social mechanisms of polarization, we specify five sufficient conditions for polarization to occur that can be evaluated using our approach. We analyze six existing network-based polarization metrics in our high-dimensional network framework through a virtual experiment and apply our proposed methodology to discussions around COVID-19 vaccines on Twitter. This work has implications for detecting polarization on social media using user-generated content, quantifying the effects of offline divides or de-polarization efforts online, and comparing community dynamics across contexts.