Speaker
Description
Online environments have become key spaces for the development and expression of extremist and cyber terrorism activity. Understanding how psychological drivers, behaviours, and group dynamics manifest in these spaces remains a significant challenge. Traditional approaches rely on manual qualitative analysis, which is resource intensive and often exposes researchers to bias, fatigue, and risk.
This research explores how attitudes, emotions, and behaviours evolve over time within discussions on Telegram channels - using data science and AI methods to classify and predict these behaviours over time. Extending prior work on large language models for extremist classification, the study incorporates a wider range of behaviours, such as misogyny, mobilisation, nostalgia, and expressions of violence.
The project further explores whether agentic large language model approaches can predict future extremist behavioural trajectories and evaluates their performance against conventional machine learning techniques, including time series models. The aim is to deepen understanding of radicalisation trajectories and approaches for classifying and predicting extremism online.
| Institutional Affiliation | University Of Bristol |
|---|