A Link Prediction Method That Can Learn from Network Dynamics

Link prediction is an important issue in Social Network Analysis area. Most of the existing link prediction methods aim to find the missing links or to predict the future links mainly based on a static network, ignoring the evolution of the network over time. This paper proposes a link prediction method that can learn from network dynamics. Using machine learning techniques, the method models the changes over time of several structural features in the network. One classifier is trained for each structural feature and the final prediction result is obtained by weighting all the classifiers. The experimental results of three real collaboration networks show that the proposed method outperforms both a traditional static method and a state-of-art dynamic method. Moreover, the experiments also show that the ability to describe network dynamics for different structural features is also different.