Research on Bipartite Network Embedding with Auxiliary Information

Abstract

Learning a low-dimensional vector representation for each node in a network is called network embedding. These learned embeddings have shown promising results on network mining tasks such as link prediction, recommendation, and node classification. Most of the published work preserve network structure only; however, there also exist rich auxiliary information (i.e., attributes, text) along with network structure in a real-world scenario. One can consider that information along with network structure to measure and enhance the capacity of representation learning methods. Many real-world applications such as user/movies (recommendation system), traders/stocks (financial system) can be demonstrated as a bipartite network (a particular class of network). In our work, we present a method named ABiNE, short for Attributed Bipartite Network Embedding, to learn the latent representations of nodes for attributed bipartite networks. We investigate and develop to incorporate both structural information, in the context of explicit, implicit relations, and attributes proximity under the framework of bipartite network embedding (BiNE). We evaluate our method by conducting experiments on a real-world dataset, i.e., MovieLens. The results have shown the improvement in link prediction, and recommendation (personalized ranking) tasks as compared to other attributed or plain network embedding methods.