GraphUC: Predictive Probability-Based Graph-Structured Data Augmentation Model for Class Imbalance Node Classification

Class imbalance in graph neural networks (GNNs) poses a significant challenge for node classification tasks, often resulting in poor performance on minority classes due to the over-representation of majority class nodes. To address this, we introduce GraphUC, a novel graph data enhancement method that leverages predictive probabilities to identify and generate hard-to-classify minority class nodes. Meanwhile, designing N-SHA synthesises new nodes by integrating neighborhood information, effectively expanding the potential space for minority classes. Experimental results on real-world datasets such as Cora and CiteSeer demonstrate that GraphUC substantially enhances GNN's classification accuracy for minority class nodes, with a maximum improvement of 30.24% in GCN performance, while maintaining accuracy for majority class nodes. This significant advancement underscores the effectiveness of GraphUC in tackling class imbalance issues in node classification.