Link prediction based on depth structure in social networks

Link prediction is an important task in social network analysis. Considering that the properties of nodes in social networks are generally inaccurate, it is more reliable and effective to use the network structure features to predict the links in the network. However, a central challenge of such methods is how to fully mine and utilize the network structure information. Here, we introduce a deep structure link prediction model (DSLP), whose idea is to integrate multiple types of community structures and multiple topology features into one probability model. We detect three types of community structures, disjoint, crisp overlap and fuzzy overlap, and then design an edge probability parameter to reflect their importance. Additionally, we propose an effective method to aggregate multiple topology features based on nodes and paths. We perform extensive experiments on artificial networks and real-world social networks to compare the proposed method with nine baseline algorithms, and the results show that our method offers higher precision than that of these well-known approaches. Finally, we discuss the method of integrating trusted node properties and feature selection.