Hybrid sampling-based contrastive learning for imbalanced node classification