Enhancing the Influence of Labels on Unlabeled Nodes in Graph Convolutional Networks

The message-passing mechanism of graph convolutional networks (i.e., GCNs) enables label information to reach more unlabeled neighbors, thereby increasing the utilization of labels. However, the additional label information does not always contribute positively to the GCN. To address this issue, we propose a new two-step framework called ELU-GCN. In the first stage, ELU-GCN conducts graph learning to learn a new graph structure (i.e., ELU-graph), which allows the additional label information to positively influence the predictions of GCN. In the second stage, we design a new graph contrastive learning on the GCN framework for representation learning by exploring the consistency and mutually exclusive information between the learned ELU graph and the original graph. Moreover, we theoretically demonstrate that the proposed method can ensure the generalization ability of GCNs. Extensive experiments validate the superiority of our method.

1. The message-passing mechanism of Graph Convolutional Networks (GCNs) allows labels to influence neighboring unlabeled nodes, but it does not guarantee that such influence is always positive.2. We propose a new objective function to learn a modified graph structure (ELU graph) that ensures label information positively contributes to GCN, and we develop an efficient optimization algorithm to achieve this.3. We introduce a point-wise contrastive constraint to fully explore the complementary and conflicting information between the original graph and the enhanced (ELU) graph.4. This will inspire new directions in graph structure learning and promote the development of better graph learning objectives.