A novel link prediction approach for scale-free networks

The link prediction problem is to predict the existence of a link between every node pair in the network based on the past observed networks arising in many practical applications such as recommender systems, information retrieval, and the marketing analysis of social networks. Here, we propose a new mathematical programming approach for predicting a future network utilizing the node degree distribution identified from historical observation of the past networks. We develop an integer programming problem for the link prediction problem, where the objective is to maximize the sum of link scores (probabilities) while respecting the node degree distribution of the networks. The performance of the proposed framework is tested on the real-life Facebook networks. The computational results show that the proposed approach can considerably improve the performance of previously published link prediction methods.