Extremist views detection: definition, annotated corpus, and baseline results

Abstract

Extremist view detection in social networks is an emerging area of research. Several attempts have been made to extremist views detection on social media. However, there is a scarcity of publicly available annotated corpora that can be used for learning and prediction. Also, there is no consensus on what should be recognized as an extremist view. In the absence of such a description, the accurate annotation of extremist views becomes a formidable task. To that end, this study has made three key contributions. Firstly, we have developed a clear understanding of extremist views by synthesizing their definitions and descriptions in the academic literature, as well as in practice. Secondly, a benchmark extremist view detection corpus (XtremeView-22) is developed. Finally, baseline experiments are performed using six machine learning techniques to evaluate their effectiveness for extremist view detection. The results show that bigrams are the most effective feature and Naive Bayes is the most effective technique to identify extremist views in social media text.