Data-driven system identification of the social network dynamics in online postings of an extremist group

Abstract

Terrorism research has begun to focus on the issue of radicalization, or the acceptance of ideological belief systems that lead toward violence. There has been particular attention paid to the role of the Internet in the exposure to and promotion of radical ideas. There is, however, minimal work that attempts to model the ways that messages are spread or how individual participation in radical on-line communities operates. In this paper, we present a stochastic linear system to represent the evolution of contribution to a sample of 126 threads in an on-line forum where individuals discuss radical belief systems. To estimate or predict the time-varying contributions of agents for given online-forum data, each agent's contribution has been modeled as a state variable. We then use the expectation-maximization (EM) algorithm to identify the model parameters including the adjacency matrix of the graph constructed among participating agents along with measurement and system uncertainty levels in online-postings. Our approach reveals the identified dynamical influences among agents in the time-varying shaping of the contribution in a datadriven fashion. We use the real-world data from online-postings to demonstrate the usefulness of our approach, and its application toward on-line radicalization.