A model to predict communications in dynamic social networks

social networks are dynamic due to continuous increases in their members, communications, and links, while these links may be lost. This study was conducted with the aim of investigating the link and communication between social network users using the centrality criterion and decision tree.

After checking the nodes in the network for each pair of unrelated nodes, some common nodes in the proximity list of these two groups were extracted as common neighbors. Analysis was performed based on common neighbors, association prediction process, and weighted common neighbors. Prediction accuracy improved. Centrality criteria were used to determine the weight of each group. New Big Data techniques were used to calculate centrality measures and store them as features of common neighbors. Personal characteristics of users were added to build complete data for training a data mining model. After modeling, the decision tree model was used to predict communication.

There was an increase in sensitivity, which indicated model power in identifying positive categories (i.e., communications) when users' characteristics were used. It means that the model could identify potential latent communications. It can be stated that users are more willing to make a relationship with users similar to them through common neighbors. Personal characteristics of users and centrality were effective in method efficiency, while removal of these properties in the learning process of the decision tree model caused a reduction in efficiency criteria.

Prediction of latent communications through social networks was promising. Better results can be obtained from further studies.