Aligning Networks by Maximizing Topological Consistency Between Neighborhoods

Network is widely used to model interactions and relationships between entities from various domains. Aligning networks refers to finding node correspondence across different networks, and could be applied to many researches including anchor link prediction on social networks, inter-network connectivity of molecular networks and information diffusion across networks. Traditionally, network alignment techniques utilize topology information, especially topological consistency to guide alignment process. In this paper, we first propose a graph structure which is the minimum unit to directly represent topological consistency and integrate topological consistency from multiple networks. We then propose a model integrating such structure on a single Kronecker product graph. In addition, our model utilizes the positive correlation between number of quadrilaterals and alignment probability which is rarely used before. We also do experiments on multiple kinds of real world datasets and baselines to prove potentials of our method.