It is not until recently that graph neural networks (GNNs) are adopted to
perform graph representation learning, among which, those based on the
aggregation of features within the neighborhood of a node achieved great
success. However, despite such achievements, GNNs illustrate defects