Topological Deep Network Embedding

Graph mining methods have been used to analyze complex network systems. However, with the increase in the complexity and dimensions of these networks, it is often required to have access to high computational processing to extract insights from networks. As such, it becomes vital to map complex graph into low dimensional vector space. Network embedding has been used to generate representation of graph in feature vector form. This brings many challenges as to what the dimension of the vector space should be, and which structure of the network has to be preserved. In this paper, we propose a semi-supervised deep learning model to generate network representation by preserving local and global structural properties of the network. We test the deep learning model on three datasets and compare it with state-of-the-art embedding methods. The results reveal effectiveness of the proposed method.