The team proposed a novel disentangled graph convolutional network named D 2-GCN to dynamically adjust the number of each node’s disentangled feature channels during training, which helps obtain ...