Online Library

This study investigated the phenomena of group polarisation with particular attention to the differences between offline and online settings. Polarisation is a process that leads people to develop extreme ideologies. Three hundred and seven participants were recruited and randomly assigned to different experimental conditions, i.e. antisocial and prosocial polarisation, within groups of 6 people composed of four confederates, participating in discussions about a social dilemma under two different circumstances: face to face and online. The degree of polarisation was assessed considering the final decisions adopted by the participants, as well as the internal dynamics characterising their final attitudes, i.e. compliance versus conversion. Results showed that online groups appeared more susceptible to polarisation and their members reported a greater degree of conformism. In particular, within online environments, the risk of being polarised, both antisocially and prosocially, increased by around 12%. Furthermore, in an online setting, a greater degree of conversion emerged only when the members decided to adopt a pro-social behaviour, while a greater degree of compliance emerged whenever they decided to adopt antisocial behaviour.

Any intervention in the violent acts of terrorist groups requires accurate differentiation among the groups themselves, which has largely been overlooked in their study beyond qualitative work. To explore the notion of terrorist group differentiation, the online communication of six violent groups were collected: Al-Nusrah Front, al-Qa’ida Central, al-Qa’ida in the Arabian Peninsula, Hamas, Islamic State of Iraq and Syria, and Taliban. All six groups embedded their ideology in digitised documents that were shared through multiple online social networks and media platforms in attempts to influence individuals to identify with their beliefs. The way these groups constructed social roles for their supporters in their ideology was proposed as a novel way to differentiate them and key term extraction was used to find important terms referenced in their communication. Experimental classification was devised to find the highest-ranking roles capable of prediction. Role terms produced high accuracy scores across experiments differentiating the groups (95%CI: 95–98%), with varying inter-group and intra-ideological differences emerging from authority-, religion-, closeness-, and conflict-based social roles. This suggests these constructs possess strong predictive potential to separate terrorist groups through nuanced expressions observed in their communication behaviour and advances our understanding of how these groups deploy harmful ideology.

This chapter explores the intersection of white and male supremacy, both of which misrepresent women as genetically and intellectually inferior and reduce them to reproductive and/or sexual functions. The white power movement historically has been characterized by sexism and misogyny, as evidenced by the movement’s attempts to retain European heritage and maintain whiteness by policing the behavior and controlling the bodies of white women. However, the influence of white supremacist discourses on physically violent manifestations of the male supremacist movement remains largely understudied. Using supervised machine learning, we compare a corpus of violent male supremacist manifestos and other multimodal content with highly influential white nationalist texts and the manifestos of violent white supremacists to identify the shared beliefs, tropes and justifications for violence deployed within.

Extremist online networks reportedly tend to use Twitter and other Social Networking Sites (SNS) in order to issue propaganda and recruitment statements. Traditional machine learning models may encounter problems when used in such a context, due to the peculiarities of microblogging sites and the manner in which these networks interact (both between themselves and with other networks). Moreover, state-of-the-art approaches have focused on non-transparent techniques that cannot be audited; so, despite the fact that they are top performing techniques, it is impossible to check if the models are actually fair. In this paper, we present a semi-supervised methodology that uses our Discriminatory Expressions algorithm for feature selection to detect expressions that are biased towards extremist content (Francisco and Castro 2020). With the help of human experts, the relevant expressions are filtered and used to retrieve further extremist content in order to iteratively provide a set of relevant and accurate expressions. These discriminatory expressions have been proved to produce less complex models that are easier to comprehend, and thus improve model transparency. In the following, we present close to 70 expressions that were discovered by using this method alongside the validation test of the algorithm in several different contexts.

The strategy of jihadist groups is based on objectives that are sometimes global. Specifically, many of these groups argue that Muslims, wherever they live, should fight for the establishment of an Islamic state or, at least, for such a state to be possible elsewhere. Therefore, taking advantage of the emergence of the Internet, they initiated an equally universal narrative strategy, with the production of a great deal of content, especially audiovisual texts. The effects of this material are known and, unfortunately, may be behind the terrorist actions of various individuals in many countries. Hence the concern of academics lies with their analyses and with the development of methodologies that can successfully deal with large amounts of multimodal information. The present research, therefore, aims to apply a quantitative procedure to the analysis of jihadist propaganda. Specifically, the authors have analysed 2,211 videos belonging to different terrorist groups, by applying an image classification algorithm. The results show that this type of approach has realistic possibilities of providing relevant information about this corpus – when realized, they may help to create automated analytical tools capable of dealing with the enormous amount of information that can be disseminated on-line.

This Thesis contributes by proposing a method for identifying users believed to be at risk of radicalisation on social media, by utilising the social media networks of already radicalised individuals and a set of indicators derived from related work on radicalisation. In addition, this Thesis provides a new to the field, in-depth analysis of the personality traits of Twitter users at risk of radicalisation and how they may differ from ordinary users. The results show that the proposed data collection and annotation scheme is able to successfully identify individuals at risk of radicalisation, yielding an inter-annotator agreement, measured by Cohen’s Kappa, of 0.83. The analysis of the predicted personality traits shows that users at risk of radicalisation have common profiles for agreeableness and conscientiousness. When comparing the predicted traits to that of ordinary, non-radical Twitter users, the predictions show a marginal difference in distribution for agreeableness, openness, and conscientiousness, indicating a certain difference in personality between the two domains.