ASBiNE: Dynamic Bipartite Network Embedding for incorporating structural and attribute information

Graph representation learning (GRL) has recently gained attention and becoming popular in research community. GRL has been proven to be extremely handy for transforming large and complex network data onto a low-dimensional vector space. It opens the door to many vector-based algorithms such as link prediction, recommendation, and classification to be effectively applied to the network data. Though many GRL algorithms exist for homogeneous (one-mode) graphs, however only a few methods exist for bipartite (two-mode) graphs. Most of the existing methods for bipartite graph representation learning mainly focus on graph topology and ignore the information available in the attributes of the nodes. In this paper, we propose a novel “Attributed and Structural Bipartite Network Embedding” (ASBiNE) method. The ASBiNE incorporates both the graph topological information concerning inter-partition and intra-partition links and attributes information by generating proximity between nodes having attribute similarities. Intermediate embeddings are generated by modeling the intra-partition links in homogeneous structural and attribute networks separately, which in the end are combined through a joint optimization framework, and final embeddings are generated. The attribute and structural information share is controlled before the joint optimization step. The proposed method is evaluated on a real-life dataset through extensive experiments. The results show that the proposed method is effective and outperforms state-of-the-art baseline embedding methods.