Group tracking on dynamic networks

This paper develops a Bayesian method for inferring the evolution of hidden groups from the signatures they leave in dynamic network data. Such methods are well established for detecting groups in static networks. The dynamic generalization is based on a Markov process model for joint group-graph evolution, which is used to produce a sequential Bayesian filter for the probabilities of the group membership hypotheses. This filter is demonstrated in a simple scenario to show how both positive information (changes in network structure) and negative information (periods of no change) may be combined to track group membership optimally.