A Framework for Predicting Community Behavior in Evolving Social Networks

Abstract

The goal of this paper is to propose a complete framework for addressing the problem of predicting communities behavior in evolving social networks. The framework encompasses all steps required for community detection, analysis and behavior prediction. Our approach is based on modeling community evolution by multidimensional time series that describe the changes of each community's properties, both structural and content-based, through time. The prediction framework is based on rule discovery upon the multidimensional time series, so that based on patterns that appear in the evolution of a community's property so far, future behavior can be predicted. Finally, exploiting the similarity between the behavior of a network's communities, their multidimensional time series will be used for community clustering. Thus, rule discovery can also incorporate global rules that appear in clusters of communities as well as on the network level, so as to discover global behavior patterns that characterize all the communities of a network.