Identifying Extremism in Text Using Deep Learning

Abstract

Various forms of terrorism have become increasingly relevant in today’s world. Consequently, the utilization of the web by various terrorist groups to spread propaganda, communicate and organize has increased. However, techniques to effectively identify such material are lacking. This chapter explores an approach which can classify any piece of text as belonging to one of four extremist groups: Sunni Islamic, Antifascist Groups, White Nationalists and Sovereign Citizens. This classification is performed by LSTM models, which will be proven to be much more effective than non-deep learning approaches. This chapter will describe the performance of various models in detail. The process of creating good quality datasets for each extremist category and the unique challenges such a task presents will also be explored.