Exponential random graph modeling for diagnosis of out-of-control signals in social network surveillance: Signal interpretation in social networks with ERG modeling

Monitoring of the social networks for detecting anomalous behavior could be vital for the system's survival. This anomalous behavior could raise from any changes in behavior or attributes of a particular individual or groups of individuals in the network and causes structural changes. Multivariate statistical process control charts are effective tools for this purpose while Exponential Random Graph Models are used to model highly interdependent data of the network. So after selecting a model for specific network, T² control charts are used for monitoring the network data to detect any anomalous behavior. Then the Mason, Tracy, and Young method is utilized for interpreting an out-of-control condition. Finally, some real-world examples are used to evaluate the performance of the proposed diagnosis approach. Since complicated dependency in a social network makes different choices in model selection for Exponential Random Graph Models and this causes various results in the evaluation study, if the impact of diagnosis result is not seen in model selection, the appropriate model will not be necessarily selected and this will affect the effectiveness of the whole system. So, in this paper for improving the performance of diagnosis, two indices are introduced and added to model selection criteria and then the appropriate model could be selected based on the decision-maker's preferences.