Bipartite Network Embedding via Effective Integration of Explicit and Implicit Relations.

Network representation learning, or network embedding, aims at mapping the nodes of the network to low-dimensional vector space, in which the learned node representations can be used for a variety of tasks, such as node classification, link prediction, and visualization. As a special class of complex networks, the bipartite network is composed of two different types of nodes in which the links only exist among different types of nodes, has important applications in the recommendation system, link prediction, and disease diagnosis. However, most existing methods for network representation learning are aimed at homogeneous networks in general, while the special properties of bipartite networks are not taken into account, such as the implicit relations (i.e., unobserved links) between nodes of the same type. In this paper, we propose a novel deep learning framework for bipartite networks, which integrates the explicit and implicit relations, while preserving the local and global structure, to learn the highly non-linear representations of nodes. Extensive experiments conducted on several real-world datasets, based on the link prediction, recommendation, and visualization, demonstrate the effectiveness of our proposed method compared with state-of-the-art network representation learning based methods.