Structure and prominence in Twitter networks centered on contentious politics

With the recent explosion in availability and use of internet social media, contentious politics is currently undergoing a shift in nature. This shift has resulted in technologically savvy groups of people that utilize these new communication resources to quickly disseminate ideas, influence, news, plans, and opinions in near real time to interested, physically disconnected communities of interest. Here, we investigate activity on Twitter during the 2011 protests and revolution in Egypt by performing an analysis of the structure of direct communications and developing a model based on these analyses. We first compare the network representing the declared friends and followers with the underlying direct communication network based on mentions and retweets, performing an evaluation using goodness of model fits, and including the appropriateness of scale-free assumptions. We also examine this underlying network with an analysis of common structural tendencies by employing exponential random graph models. The network structure of locally focused Twitter communities surrounding contentious political networks tend to have a relatively high density, and a propensity for mutual, transitive, and k-star structures, suggesting that both diffusion of information and back-and-forth communication take place frequently in such networks. More global Twitter networks possess properties corroborative of scale-freeness; however, even locally bounded networks show an evidence of preferential attachment. Based on these observations, we introduce a model to automatically ascertain prominent Twitter users within networks centered on contentious politics. We employ a form of the alpha centrality, which can capably take into account directionality and exogenous prominence, and has ties to non-conservative diffusion. Results indicate a model successful at automatically identifying users that are active and prominent within a given community that agrees well with heuristics and is comparable to other similar models. The model also has advantages over others based on its tunability and robustness in the presence of incomplete data.