Analyzing violent attacker texts: examining ideologies through machine learning-assisted topic modeling

Abstract

ABSTRACT The rise in extremist violence is a global concern, prompting advancements in threat assessment. This study applies machine learning to dissect publicly available texts from violent attackers with white supremacist, incel, and political extremist ideologies, as well as from those driven by personal grievances. Through natural language processing, the research identifies specific linguistic patterns and themes inherent to each group and explores the intersections among them. It uncovers shared narratives of disenfranchisement and aggression alongside unique ideological signifiers. Additionally, it identifies themes unique to each group, offering a clear thematic distinction between ideologies. These insights contribute to a more accurate identification of radicalization indicators, supporting the development of intervention strategies tailored to the ideological nuances of each group. Enhanced threat assessment tools informed by this research can improve the precision of prevention efforts by law enforcement and mental health professionals, reflecting the complex realities of extremist motivations and actions.