Incorporating label and attribute information for enhanced network representation learning

Recently, network representation learning has shown its super performance in many network analysis tasks. Traditional network representation learning algorithms mainly mine unilateral network structure information. However, as we move to the age of big data, we can obtain rich auxiliary information from many real-world networks. In this paper, we propose an enhanced network representation learning algorithm by incorporating label and attribute Information. Inspired by natural language model, we propose two novel models: label enhanced attribute information representing model and label enhanced structure information representing model. Then, through parameter sharing strategy and jointly training based on a unified optimization objective function, we can obtain representations of nodes in network preserving network structure, attribute information and label information simultaneously. Experimental results demonstrate that representation vectors obtained by our model outperform state-of-art network representation learning methods on two real-world networks.