Impact of Intergroup Interactions on Polarization

Polarization is an important driver of conflict and social unrest. Existing research suggests that social media platforms such as Twitter may have become more polarized over time. This necessitates the need to systematically understand the temporal trends in polarization and examine the effect of interventions, such as intergroup contact, to mitigate it. One of the main challenges towards this goal is conceptualizing and estimating polarization. Towards this, we introduce a new measure for an individual's group conformity (and polarization) based on contextualized embeddings of the tweet text that can meaningfully capture different dimensions of linguistic polarization. We focus on religious polarization in the South Asian context— where religion is a salient social division. We use character-sequence-based machine learning models to infer the religious identities of nearly 700,000 Indian Twitter users engaging in COVID-19-related discourse during 2020. We then use a meta-learning framework to examine the effect of intergroup interactions on polarization. While exposure to diverse viewpoints may reduce polarization, it can also have a backfire effect and exacerbate polarization when the discussion is adversarial. We investigate the heterogeneities in the treatment effect on an individual's group conformity in light of communal, political, and socio-economic events and find that while intergroup interactions reduce polarization in general, the effects are sensitive to the context. Our findings have important implications for understanding the dynamics of religious polarization, especially under the influence of intergroup interactions, and can help inform policies to mitigate it and foster healthier social media ecosystems.