Tracing temporal communities and event prediction in dynamic social networks

Online social networks (OSNs) represent the communication structure among users in the network. This communication structure is separated into segments of highly communicative density in comparison with the rest of the network as communities. The structure of a community may change over time due to changes in the relationships between its members or connection with other communities. These changes are known as community events. However, in this paper, we present the event predicting in dynamic social network (EPDSN) method to trace temporal communities. For this purpose, we propose a new definition of events for OSNs and identify types of events in compliance with the reality of OSNs. Features and events of the community during the previous snapshot are used as the input of the classifier into the learning model. We train the EPDSN method based on real-world Facebook, Wikipedia, and an OSN at the University of California datasets. This method uses the non-overlapping snapshots to keep the reality of the event prediction for the next snapshot and deploys snapshots with the same length to preserve its generality. The experimental results confirm that the prediction of the events for a community in dynamic OSNs can be achieved with high accuracy.