Speaker
Description
This study examines rhetorical strategies used by extremists on social media and behavioral leakage signals that may accompany escalation toward violence. Simple language-based threat identification methods are limited by the fact that online extremist communities are saturated with inflammatory language, idle threats, and unserious discussions, creating a sea of false-positives that obscure genuine threats. Using a sample of over 200,000 social media posts (on the X platform) as a foundation, we analyze posts from three subgroups of January 6th capitol rioters (Passive, Active, and Violent: those arrested for violent action on Jan 6th). The analysis explores latent language differences in how subgroups within a single ideological movement communicate and interact with social media, all with the intention to identify predictors of violent action. The exploration of rhetorical and behavioral features within subgroups will be conducted using unsupervised machine learning methods, including Structural Topic Modeling (STM), Latent Dirichlet Allocation (LDA), and Correlated Topic Modeling (CTM). Alongside the latent features identified by this exploration, we will use predictors modeled after the radicalization signals described in 3N theory, Significance Quest theory, and Identity Fusion to perform a supervised prediction of subgroup membership on a user-by-user basis.
| Institutional Affiliation | University of North Florida |
|---|