Structure, Attribute and Homophily Preserved Social Network Embedding

Abstract

Network embedding is to map nodes in a network into low-dimensional vector representations such that the information conveyed by the original network can be effectively captured. We hold that a social network mainly contains three types of information: network structure, node attributes, and their correlation called homophily. All of these information could be potentially helpful in learning an informative network representation. However, most existing network embedding methods only consider one or two types of these information, which are possibly leading to generate unsatisfactory representation. In this paper, we propose a novel algorithm called Structure, Attribute, and Homophily Preserved (SAHP), which jointly exploits the aforementioned three information for learning desirable network representation. And we design a joint optimization framework to embed the three information into a consistent subspace where the interplay between them is captured toward learning optimal network representations. Experiments conducted on three real-world social networks demonstrate that the proposed algorithm SAHP outperforms the state-of-the-art network embedding methods.