Weighted Graph Classification by Self-Aligned Graph Convolutional Networks Using Self-Generated Structural Features

Directed weighted graphs are important graph data. The weights and directions of the edges carry rich information which can be utilized in many areas. For instance, in a cashflow network, the direction and amount of a transfer can be used to detect social ties or criminal organizations. Hence it is important to study the weighted graph classification problems. In this paper, we present a graph classification algorithm called Self-Aligned graph convolutional network (SA-GCN) for weighted graph classification. SA-GCN first normalizes a given graph so that graphs are trimmed and aligned in correspondence. Following that structural features are extracted from the edge weights and graph structures. And finally the model is trained in an adversarial way to make the model more robust. Experiments on benchmark datasets showed that the proposed model could achieve competitive results and outperformed some popular state-of-the-art graph classification methods.