Dynamic bipartite network model based on structure and preference features

Based on the complex network, the relationship in the real complex system can be modeled, and the bipartite network is a special complex network, which can describe the complex system containing two kinds of objects. Although existing bipartite networks can model complex systems, conventional methods are restricted to a couple of limitations. (1) The dynamic interaction between nodes cannot be described over time. (2) The implicit features of nodes in the network cannot be effectively mined. Based on these, this paper proposes a dynamic bipartite network model (DBN) to model the dynamic interaction between two types of objects in real complex systems, and mine the structure features and preference features of nodes in the network. First, the dynamic interaction between two types of objects in a complex system is modeled as a dynamic bipartite network, which can reflect the interaction between objects in each time slice. Then, the structure features and preference features of each time slice are mined based on the dynamic bipartite network, where the structure features reflect the dynamic structural changes of the nodes, and the preference features reflect the potential preferences of the nodes. Finally, the features of each time slice are fused and input into the gate recurrent unit model to predict the interaction between nodes. Extensive experiments are performed on a large-scale real complex system. The results show that DBN significantly outperforms state-of-the-art prediction methods in terms of multiple evaluation metrics.