Experimental results on several databases show that the introduced augmentation framework improves the performance of existing graph neural networks on both semi-supervised node classification and ...