Predictive Analysis of Temporal and Overlapping Community Structures in Social Media

We perform a study on community evolution analysis and prediction. We employ three overlapping community detection (OCD) algorithms from literature to the case of time-evolving networks including social, email communication and co-authorship networks. Group evolution discovery (GED) technique is applied to track the identified communities. We compare structural properties of OCD algorithms and investigate most persistent communities over time. Furthermore, static and temporal features of a community are applied to build a logistic classifier for community evolution prediction (CEP). Results indicate that size of communities can be considered for predicting four events which happen to a community such as split, merge, dissolve and survive. Moreover, importance of features like cohesion, density, assortative degree mixing and centralities depend on intended events and the definition of communities.