Link Prediction of Heterogeneous Information Networks Based on Frequent Subgraph Evolution

The problem of link prediction in heterogeneous information networks has been widely studied in recent years. It is essential to grasp the evolution law for both static information networks and dynamic information networks. However, most existing works only focus on the global topology, ignoring the impact of network microstructure on network evolution. In this paper, we present a novel link prediction method called LP-FSE (link prediction based on frequent subgraph evolution) of heterogeneous information networks. Different from traditional methods, LP-FSE makes full use of the microstructure and dynamic characteristics of networks to predict links. On one hand, frequent subgraphs are mined in heterogeneous information networks, and accord to the amount of different frequent subgraphs to predict the trend of frequent subgraphs. On the other hand, the evolution process of approximate subgraph tending to frequent subgraph is studied. Then, an approximate subgraph matching method based on matrix sliding algorithm is proposed. Experiments demonstrate the effectiveness and the efficiency of our proposed method.