A Multi-Dimensional Framework to Analyze Group Behavior Based on Political Polarization

The recent wave of elections won by right-wing worldwide brings up increased discussions biased by political polarization, including in social media. Social media data enables the investigation of the contexts where political polarization occurs, enabling to derive insights into how it affects human behavior. Related work has shown how computing techniques can be leveraged to understand political polarization in restricted scenarios, but the complexity of this behavior can be better understood when considered from different viewpoints. This article describes a multi-dimensional analysis framework to study the behavior of groups on Twitter in politically polarized scenarios. It can be applied to various themes where groups display stances that can be politically biased, and it aggregates a wide range of computational techniques in an innovative way to provide rich insights. The framework includes guidelines and techniques to: a) collect data on Twitter to represent the groups; b) automatically infer the political leaning of users; c) derive topological properties of the groups' social network and analyze political influence; d) identify topics representing concerns at coarse and fine-grained granularity levels using a hybrid topic modeling approach; e) identify psychological aspects based on linguistic cues, and f) analyze the sources of information disseminated by the groups. Applying the framework in two case studies related to COVID revealed patterns of behavior common to ideologies, showing the generalization of the analysis in politically polarized groups in different subjects.