Abstract:
Most of the existing link prediction methods are aimed at homogenous networks without considering that the real networks are heterogeneous networks with different node or edge properties. This kind of methods cannot make full use of the topological structure information of different types of nodes or edges. Under this circumstance, this paper proposes a link prediction method based on heterogeneous phantom features, which divides users in the network by gender differences as node types, and constructs heterogeneous phantom features distinguishing node types for link prediction in heterogeneous networks. On this basis, a link prediction algorithm that combines the characteristics of the homogeneous phantom and the heterogeneous phantom is proposed. Compared with the performance of the existing prediction method on the real data set, the AUC value is increased by 17.0%~23.1%, and the precision value is increased by 7.6%~20.1%. This method can be used in online social networks to explore the role and influence of user gender differences on interpersonal communication, and then to analyze the evolutionary dynamics of heterogeneous networks.