Bipartite graph capsule network

Abstract Graphs have been widely adopted in various fields, where many graph models are developed. Most of previous research focuses on unipartite or homogeneous analysis. In this graphs, the relationships between same type entities preserved graphs. Meanwhile, bipartite graphs that model complex among different with vertices partitioned into two disjoint sets, becoming increasing popular and ubiquitous real life applications. Though several classification methods homogenous proposed by using kernel method, neural network, etc. However, these unable to effectively capture hidden information paper, we propose first graph-based capsule namely Bipartite Capsule Graph Neural Network (BCGNN), for task. BCGNN exploits network obtains constructing one-mode projection. Extensive experiments conducted real-world datasets demonstrate effectiveness our method.