User Identity Modeling to Characterize Communication Patterns of Domestic Extremists Behavior on Social Media.

Years of online manifestos and mass killings by Domestic Violent Extremists (DVEs) in the U.S. and other nations indicate a grave global crisis. DVEs employ online propaganda to maximize their influence during societal crises and become the conduits to amplifying hate, organizing, and mobilizing to possibly influence non-extremists. User behavior modeling provides a promising approach to characterize behavioral patterns of DVEs and model the evolving identities of users. However, empirical research falls short to create explanatory user behavior modeling techniques for learning the classes of DVE user identity. Specifically, existing techniques lack measures to make sense of DVEs’ online behavior relating to the well-known factors such as social relationships and beliefs that contribute to shaping one’s identity. Guided by social science theories, primarily Social Identity Theory (SIT), this study proposes an explanatory and scalable user modeling approach that leverages unsupervised machine learning to analyze and model the identity classes of users. We model the identity classes of DVEs through a set of derived behavioral attributes–sociability, reach, and subjectivity from user profile metadata available in streaming social media data that helps capture the factors of social relations and attitudes partially but rapidly at scale and in near real-time. Using the data of Twitter conversations leading up to the 2021 January 6th U.S. Capitol attack, this study presents novel insights on how DVEs identities can be discovered and present better distinct analyses to understand their differences.