ANE: Network Embedding via Adversarial Autoencoders

Network embedding is an important method to learn low-dimensional representations of vertexes in network, whose goal is to capture and preserve the highly non-linear network structures. Here, we propose an Adversarial autoencoders based Network Embedding method (ANE for short), which utilizes the rencently proposed adversarial autoencoders to perform variational inference by matching the aggregated posterior of low-dimensional representations of vertexes with an arbitraray prior distribution. This framework introduces adversarial regularization to autoencoders. And it is able to attaches the latent representations of similar vertexes to each other and thus prevents the manifold fracturing problem that is typically encountered in the embeddings learnt by the autoencoders. Experiments demonstrate the effictiveness of ANE on link prediction and multi-label classification on three real-world information networks.