The paper introduces the Heterogeneous Graph Transformer (HGT), a novel architecture designed to model Web-scale heterogeneous graphs. HGT addresses the limitations of existing graph neural networks (GNNs) by incorporating node- and edge-type dependent parameters to capture heterogeneous attention over edges. It also introduces the relative temporal encoding (RTE) technique to handle dynamic heterogeneous graphs, enabling the model to capture structural dependencies with arbitrary durations. To efficiently train on large-scale graphs, HGT employs the heterogeneous mini-batch graph sampling algorithm (HGSampling). Extensive experiments on the Open Academic Graph (OAG) dataset, which contains 179 million nodes and 2 billion edges, demonstrate that HGT consistently outperforms state-of-the-art GNN baselines by 9%–21% on various downstream tasks. The paper also includes an ablation study and a case study to highlight the effectiveness of HGT's key components.The paper introduces the Heterogeneous Graph Transformer (HGT), a novel architecture designed to model Web-scale heterogeneous graphs. HGT addresses the limitations of existing graph neural networks (GNNs) by incorporating node- and edge-type dependent parameters to capture heterogeneous attention over edges. It also introduces the relative temporal encoding (RTE) technique to handle dynamic heterogeneous graphs, enabling the model to capture structural dependencies with arbitrary durations. To efficiently train on large-scale graphs, HGT employs the heterogeneous mini-batch graph sampling algorithm (HGSampling). Extensive experiments on the Open Academic Graph (OAG) dataset, which contains 179 million nodes and 2 billion edges, demonstrate that HGT consistently outperforms state-of-the-art GNN baselines by 9%–21% on various downstream tasks. The paper also includes an ablation study and a case study to highlight the effectiveness of HGT's key components.