Mining Pro-ISIS Radicalisation Signals from Social Media Users

Abstract

The emergence and actions of the so-called Islamic State of Iraq and the Levant (ISIL/ISIS) has received widespread news coverage across the World, largely due to their capture of large swathes of land across Syria and Iraq, and the publishing of execution and propaganda videos. Enticed by such material published on social media and attracted to the cause of ISIS, there have been numerous reports of individuals from European countries (the United Kingdom and France in particular) moving to Syria and joining ISIS. In this paper our aim to understand what happens to Europe-based Twitter users before, during, and after they exhibit pro-ISIS behaviour (i.e. using pro-ISIS terms, sharing content from pro-ISIS accounts), characterising such behaviour as radicalisation signals. We adopt a data-mining oriented approach to computationally determine time points of activation (i.e. when users begin to adopt pro-ISIS behaviour), characterise divergent behaviour (both lexically and socially), and quantify influence dynamics as pro-ISIS terms are adopted. Our findings show that: (i) of 154K users examined only 727 exhibited signs of pro-ISIS behaviour and the vast majority of those 727 users became \emph{activated} with such behaviour during the summer of 2014 when ISIS shared many beheading videos online; (ii) users exhibit significant behaviour divergence around the time of their activation, and; (iii) social homophily has a strong bearing on the diffusion process of pro-ISIS terms through Twitter.