Adversarial Representation Mechanism Learning for Network Embedding

Network embedding which is to learn a low dimensional representation of nodes in a network has been used in many network analysis tasks. Recently some Generative Adversarial Networks (GAN) based network embedding methods have been proposed. These methods typically use GAN to force the network embedding results to follow a priori distribution (e.g. Gaussian distribution in most cases), which do not make full use of the essential advantage of adversarial learning. To address this problem, we propose a novel adversarial learning framework with three players, which applies the adversarial learning strategy on the representation mechanism. In the new framework, besides the discriminator, there are two other players, named positive sample generator and competitor, which aim to learn two different representation mechanisms. These two players compete with each other to improve their representation mechanisms. Furthermore, to design a competitive competitor, we use the same framework as the positive sample generator but take as input a fake network generated by a designed perturbation strategy. The experimental results show the superior performance of the proposed approach over the existing methods (including the state-of-the-art of embedding method DGI) on a variety of network analysis tasks including node clustering, node classification, link prediction and visualization.