PRISM: Predicting Radicalization Events in Social Media User Timelines
At a glance
- Project leader : Prof. Dr. Mark Cieliebak
- Deputy of project leader : Dr. Don Tuggener
- Project team : Pius von Däniken
- Project budget : CHF 1'497'093
- Project status : ongoing
- Funding partner : Federal government (Bundesamt für Rüstung armasuisse / armasuisse W+T - Wissenschaft und Technologie)
- Project partner : Bundesamt für Rüstung armasuisse / armasuisse W+T - Wissenschaft und Technologie
- Contact person : Mark Cieliebak
The PRISM project focuses on detecting radicalization events in Social Media networks. Overall, we are interested in unveiling the mechanics that lead to the event of extremist ideology being transferred and incorporated into a social media user’s world view. Specifically, the proposed project aims to identify emerging and ongoing radicalization events in a user’s social media timeline.
Different from related work on online radicalization, we do not assume that a user is already radicalized, but rather aim to identify developments and features that indicate that a user is (prone to) undergoing radicalization before or as it happens. Specifically, we are interested in the observable features of a user’s social media behaviour that accompany radicalization. These features can be broadly grouped into two categories:
- a. User-generated content: Topic detection and shifts, shifts in polarity, sentiment, and tonality towards target entities of interest in the user’s own tweets.
- b. Social network behaviour: (Re)tweet behaviour, tweets reacted to and the type of reaction, social graph topology (following/followers/friends).
For the user content analysis, we will apply methods such as topic detection and tracking, stance detection, sentiment analysis, and opinion mining. We will additionally explore models of argumentation to understand what narrative a user might be following. Furthermore, persuasion is an important factor in radicalization, and we will apply models to detect and analyse persuasive dialogues a user might be involved in. To tackle feature group b), we will leverage various network related analyses, metrics, and statistics, such as post frequency and changes in the social graph structure. Finally, we will combine these features in a predictive machine learning model.