Keep It Simple: Graph Autoencoders Without Graph Convolutional Networks

Graph autoencoders (AE) and variational autoencoders (V AE) recently emerged as powerful node embedding methods, with promising perform ances on challenging tasks such as link prediction and node clustering. Graph AE, V AE and most of their extensions rely on graph convolutional networks (G CN) to learn vector space representations of nodes. In this paper, we propose to replace the GCN encoder by a simple linear model w.r.t. the adjacency matrix of the graph. For the two aforementioned tasks, we empirically show that this app roach consistently reaches competitive performances w.r.t. GCN-based models for numerous real-world graphs, including the widely used Cora, Citeseer and P ubmed citation networks that became the de facto benchmark datasets for evaluating graph AE and V AE. This result questions the relevance of repeatedly usin g these three datasets to compare complex graph AE and V AE models. It also emphasizes t he effectiveness of simple node encoding schemes for many real-world applica tions.