Moreover, based on a novel heterogeneous graph representation of scheduling states, a heterogeneous-graph-neural-network-based architecture is proposed to capture complex relationships among ...