Adversarial Graph Disentanglement With Component-Specific Aggregation

A real-world graph has a complex topological structure, which is often formed by the interaction of different latent factors. Disentanglement of these latent factors can effectively improve the robustness and expressiveness of the node representation of a graph. However, most existing methods lack consideration of the intrinsic differences in relations between nodes caused by factor entanglement. In this article, we propose an adversarial disentangled graph convolutional network (ADGCN) for disentangled graph representation learning. To begin with, we point out two aspects of graph disentanglement that need to be considered, i.e., microdisentanglement and macrodisentanglement. For them, a component-specific aggregation approach is proposed to achieve microdisentanglement by inferring latent components that caused the links between nodes. On the basis of microdisentanglement, we further propose a macrodisentanglement adversarial regularizer to improve the separability among component distributions, thus restricting the interdependence among components. In addition, to reveal the topological graph structure, a diversity-preserving node sampling approach is proposed, by which the graph structure can be progressively refined in a way of local structure awareness. The experimental results on various real-world graph data verify that our ADGCN obtains more favorable performance over currently available alternatives.