Understanding the (de)polarized social media

Political polarization poses a serious societal challenge around the globe, which has only grown more pronounced with the rise of digital communication. As increasingly central venues for political discourse, social media platforms provide unprecedented opportunities for studying polarization dynamics through unobtrusive observations. Building on prior research that assessed whether social media is polarized and to what extent, this thesis delves into polarized debates to understand how polarization persists and deepens in online discussions. A case study of climate change discussions on Twitter points to information spreading as a crucial reinforcing mechanism of polarization: not only does content circulate primarily within like-minded groups, but the most viral themes also function to enhance ingroup bonds and preclude outgroup engagement. In addition to the empirical analysis, a technical framework is presented for automatically decomposing layers of polarized debates from traces of information diffusion, while its applicability to social media data is systematically evaluated. The second part of this thesis addresses the more challenging question of how polarization can be mitigated. Despite the success of various depolarization experiments, empirical investigation into depolarization on social media remains scarce. To fill this gap, this thesis examines depolarization in real-world settings following cross-party interactions or external threats. The results show limited evidence of depolarization after a cross-party interaction in Reddit political discussions. Even in the face of a substantial external threat, political debates on Twitter were selectively depolarized, with consensus reached among a subset of actors on a subset of arguments. The contribution of this thesis is two-fold. In rendering a realistic depiction of (de)polarized social media discourse, the findings reveal novel insights into the depth of polarization in the system and its striking resilience to depolarization. Through a critical review of the research methods employed, the thesis presents a broader reflection on the merits and limits of observational analysis in computational social science, while underscoring the value of methodological triangulation and integration. In doing so, it not only advances present understandings of political polarization, but also informs future inquiries into the phenomenon.