Abstract: Existing message passing-based and transformer-based graph neural networks (GNNs) cannot satisfy requirements for learning representative graph embeddings due to restricted receptive fields, ...