Toward Community Dynamic through Interactions Prediction in Complex Networks

Until recently all the works done on community detection in complex networks have only consider static networks: a snapshot of the network is taken at a particular time. The communities are then computed on that constructed network. Because real networks are dynamic by nature, investigations on community detection in dynamic networks have started these last years. One problem actually unexplored in community dynamic is the prediction: knowing the evolution of the network until the time-step t, can we predict the communities at the time-step t+1? In this paper, we propose a general approach for communities prediction based on a machine learning model predicting interaction in social networks. In fact, we believe that if one is able to predict the structure of the network with a high precision, then one just need to compute the communities on this predicted network to have the prediction of the community structure. Evaluation on real datasets (DBLP and Facebook walls) shows the feasibility of the approach.