Heterogeneous graphs organize data with nodes and edges, and have been widely used in various graph-centric applications.

Classic Graph Convolutional Networks (GCNs) often learn node representation holistically, which would ignore the distinct impacts from different neighbors when aggregating their features to update a ...