Description of original award (Fiscal Year 2021, $1,126,264)
The proposed research is a secondary data analysis of individual-level social media data. The goal of this research is to better understand the degree to which linguistic risk factors can be used to predict violent extremist outcomes at the individual level. Furthermore, it seeks to identify the degree to which these linguistic risk factors vary between actor ideologies, between actor types, and within the same actor over time.
Specifically, the proposed research will develop a retrospective, longitudinal database of regular social media users. The project will identify and collect data from individuals who posted expressions of ideology online, including at least (1) 200 violent extremist offenders, (2) 200 matched nonviolent extremists, (3) 200 non-extremist, partisan social media users who posted partisan social media content and who will serve as controls to extremists to identify transitions to extremism, (4) 100 far-right extremist offenders, (5) 100 leftwing extremist offenders, (6) 100 religious extremist offenders, (7) 100 lone-actor offenders, and (8) 100 group-based offenders. This overall sample will cover the spectrum of violent extremist actors; including lone actors and cell-based actors who operated within the United States since 2000, individuals who departed the United States to join foreign insurgencies, and group-based opportunistic actors (e.g., those who participated in the January 6 Capitol riot) who committed extremist violence during a larger public event. This data will then be analyzed using a psychologically validated linguistic text analysis program (Linguistic Inquiry Word Count; LIWC) to identify domestic extremism and violent extremist plots. The proposed research will include downloading and cleaning longitudinal social media language data from thousands of known violent and nonviolent (matched) extremist social media users through the Twitter Academic API and other social media sites to (a) validate pilot findings through a new dataset of violent extremist offenders in the United States, (b) examine different LIWC characteristics across actor ideologies and actor types, and (c) detect the risk of transitioning to extremism using an analysis of longitudinal linguistic behavior.
Research Questions include:
1. To what degree does LIVE differentiate violent from nonviolent extremist individuals?
2. Do similar or unique linguistic characteristics assessed with LIVE differentiate violent and nonviolent extremists when comparing the type of actor (lone actors vs. group-based) and ideology (far-right vs. far-left)?
3. To what extent does LIVE predict transitions from non-extremist to extremist posting on social media?
"Note: This project contains a research and/or development component, as defined in applicable law," and complies with Part 200 Uniform Requirements - 2 CFR 200.210(a)(14). CA/NCF