Computational and formal methods in terrorism research
Thursday June 20, 15.30 - 17.00
Session 3, Auditorium 6, Eilert Sundt building
Chair: Emily Corner
- Vulnerability assessment based on probability modelling - Dianne van Hemert, Tony van Vliet, Bob van der Vecht
- Analysing radicalisation through social networks, a cooperative game theoretic approach - H.W. Meerveld and R.H.A. Lindelauf
- Tracking the Mindset Changes associated with Online Radicalization through Linguistic Growth Curves of Far-Right Extremist Forums - Shuki Cohen
Vulnerability assessment based on probability modelling
Dianne van Hemert, Tony van Vliet, and Bob van der Vecht, TNO, The Netherlands
We present a new European vulnerability assessment tool for violent radicalisation, that is based on a Bayesian probability model. The tool is designed for multidisciplinary teams, for example consisting of police, municipalities, social work, and schools. It aims to help teams to assess how vulnerable an individual is to radicalise towards using violence, and which information is needed to increase the reliability of the assessment. The 80 indicators in the tool were linked to different (professional) perspectives based on two types of probability elicitation. First, radicalisation experts identified for each indicator how often it reflects vulnerability on an aspect of radicalisation; for example, of 100 scrutinised individuals posting radical messages on the internet, how many are vulnerable for preparing a violent act? Second, professionals from the field are requested to indicate of individual cases containing several indicators how vulnerable they think the individuals are. Together, these probability estimates provide input for a model that underlies the vulnerability assessment tool. These will, in time, be replaced by data obtained from vulnerability assessments obtained by use of the tool making it a more data-driven tool than many in the field.
Analysing radicalisation through social networks, a cooperative game theoretic approach
H.W. (Herwin) Meerveld and R.H.A. (Roy) Lindelauf, Royal Netherlands Air Force / Netherlands Defence Academy
During the last decade Europe has been the unfortunate witness of multiple terrorist attacks. These attacks were conducted by networks of individuals that shared certain radical beliefs. The radicalisation process and how to prevent or subdue it clearly is an important topic to European intelligence and security agencies. It is well known that the structure of social networks facilitates or inhibits the spread of ideas and information (Granovetter, 1973; Shakarian et al., 2015). Henceforth it is expected that the structural position of individuals in the social network is a strong determinant of their expected future level of radicalization. A possible de-radicalisation strategy should therefore take the aspect of diffusion of ideas through social networks into account. This paper introduces a first radicalisation model (RM) wherein the Shapley value serves as an input for the Linear Threshold Model (LT). The Shapley value is used to quantify someone’s influence on the radicalisation of others. The LT models the diffusion of this influence. This RM provides new insights on the importance of certain players and connections in a social network regarding the radicalisation of individuals. It can therefore be used to better allocate scarce intelligence resources, quantifying people’s vulnerability to radical ideas. Furthermore, the model is robust enough to neutralise the fact that intelligence agencies can lack information.
Tracking the Mindset Changes associated with Online Radicalization through Linguistic Growth Curves of Far-Right Extremist Forums
Shuki J. Cohen, John Jay College of Criminal Justice
Despite considerable anecdotal evidence and intuitive appeal, the process by which online engagement radicalizes individuals to espouse hateful and violent extremist ideologies is still insufficiently understood. This study uses linguistic methods to gauge changes in cognitive and emotional information processing among users of a large, US-focused right-wing extremist online forum. Mixed-Effects Linear and Quasi-Linear models were applied to a 1.8 Million words corpus, spanning several years of forum activity, to estimate the commonalities in the individual linguistic (and arguably mental) changes with their tenure on the forum. Consistent with models suggesting an ‘addictive’ quality to social media, the results showed an overall tendency of users to write, on average, increasingly longer posts. However, the temporal changes in the content of the posts was more complex: Consistent with several models of radicalization, the tendency to use absolutist, overgeneral and overconfident words (as a proxy of fanaticism) increased with time, while emotion words decreased – suggesting a radicalization trajectory consisting of growing fanatic detachment (presenting as ‘factual’ bigotry) rather than fanatic rage. This unique glimpse into the time course of online radicalization and the nature of its content may inform both timing and substance of counter-messaging campaigns to mitigate radicalization to violent extremism.