On July 26-30, 2017, the 1st North American Social Networks (NASN) Conference, a regional conference of the International Network of Social Network Analysis (INSNA), hosted a conference in Washington, D.C. The NASN conference provides an interdisciplinary venue for social scientists, mathematicians, computer scientists, ethnologists, epidemiologists, organizational theorists, public health experts, and others to present current work in the area of social networks.
The conference’s prominent keynote speaker was David Lazer, Distinguished Professor of Political Science and Computer and Information Science, Northeastern University, and Co-Director, NULab for Texts, Maps, and Networks.
Joe Shaheen, Computational Social Science PhD student within the Department of Computational and Data Sciences, was invited to present his paper “Combining Social Network Analysis with Agent-based Modeling to Reproduce an ISIS Social Media Network” at this conference. The paper focused on using social effects to reproduce a collected friend and follow network from social media using bottom-up growth.
An agent-based model is developed to understand the behaviors and rule-sets that generate social media networks. Simple rules are used to synthetically generate a backcloth (friend/follow) network collected using the Twitter API. The Twitter network was collected using seed accounts for known terrorist propaganda accounts. Model parameter adjustments were made to reproduce the collected net-work’s summary statistics and stylized specifics such as average degree, clustering, community size and distribution, as well as general structural metrics. An approximate network was produced in line with the general properties of our collected data. In this paper, we present our findings on the reproduction of a social media network with a focus on testing similarity of summary statistics and structural properties. We find that it is possible to generate a social media network utilizing a few simple rules and a unique time-rule which sets varying interaction rates on entry of new nodes in comparison to existing node activity. We call this time rule mechanism a coupled rule-set. We also present weaknesses in our reproduction and propose an extension of the model for future work which could better reproduce more exact network properties.