Link prediction in dynamical networks based on mutual information

The critical issue of link prediction is to measure the existence probability of new links between nodes with the known node properties and observed structural features. The previous researches on complex network are conducted based on the oversimplified assumption that the network structure is static, while the temporal dimension has great impacts on structure and dynamic of real systems, which limits the performance of state-of-the-art methods. Inspired by this, we proposed moving average mutual information (MA_MI) algorithm that combines mutual information with moving average model. This method not only considers the information of common neighbors of nodes, but also describes the evolution pattern of network with historical information. Experimental results on four real-world networks show that MA_MI outperforms the traditional methods and achieves the higher precision.