This paper introduces a novel heterogeneous graph neural network (HAN) that leverages hierarchical attention mechanisms to handle heterogeneous graphs, which contain different types of nodes and links. The proposed HAN addresses the challenges posed by the heterogeneity and rich semantic information in heterogeneous graphs. Specifically, it includes node-level attention and semantic-level attention. Node-level attention learns the importance between a node and its meta-path-based neighbors, while semantic-level attention learns the importance of different meta-paths. The model aggregates features from meta-path-based neighbors in a hierarchical manner, considering both node-level and semantic-level attentions. Extensive experiments on real-world heterogeneous graphs demonstrate the superior performance and interpretability of the proposed HAN compared to state-of-the-art models. The contributions of the work include the first attempt to apply attention mechanisms to heterogeneous graph neural networks, the development of a highly efficient model with linear complexity, and the demonstration of good interpretability for heterogeneous graph analysis.This paper introduces a novel heterogeneous graph neural network (HAN) that leverages hierarchical attention mechanisms to handle heterogeneous graphs, which contain different types of nodes and links. The proposed HAN addresses the challenges posed by the heterogeneity and rich semantic information in heterogeneous graphs. Specifically, it includes node-level attention and semantic-level attention. Node-level attention learns the importance between a node and its meta-path-based neighbors, while semantic-level attention learns the importance of different meta-paths. The model aggregates features from meta-path-based neighbors in a hierarchical manner, considering both node-level and semantic-level attentions. Extensive experiments on real-world heterogeneous graphs demonstrate the superior performance and interpretability of the proposed HAN compared to state-of-the-art models. The contributions of the work include the first attempt to apply attention mechanisms to heterogeneous graph neural networks, the development of a highly efficient model with linear complexity, and the demonstration of good interpretability for heterogeneous graph analysis.