<p>The importance of node ranking in heterogeneous networks is crucial for uncovering key nodes, such as opinion leaders in social networks or critical institutions in financial systems. However, existing methods face several limitations, particularly in measuring node importance accurately in networks with diverse node types. Traditional approaches often overlook node heterogeneity, like treating all nodes as a single type, while meta-path-based methods in heterogeneous networks require domain knowledge and lack scalability. To address these two limitations, we propose Heterogeneous-Homogeneous Graph Neural Network (HHGNN), a novel framework that leverages the extraction of homogeneous subgraphs from heterogeneous networks and learns representations based on edge information within these subgraphs. HHGNN introduces a novel multi-order contextual adjacency matrix (MO-CAM) that captures interaction strengths across varying context path lengths, combined with a cross-order message passing mechanism. By leveraging MO-CAM, the edge-level self-attention dynamically adjusts the influence of different node interactions, enabling the model to adaptively balance low-order and high-order path information. Specifically, MO-CAM provides the necessary basis for the model's dynamic adjustment of edge-level attention, enhancing its ability to accurately reflect the importance of various node interactions. This unique combination ensures accurate node influence ranking while maintaining scalability and flexibility in heterogeneous networks. Comprehensive experiments based on real-world datasets are conducted. Experiments results demonstrate that HHGNN outperforms existing baseline models, achieving superior node ranking performance. Our method provides a scalable and effective solution for node influence ranking in heterogeneous networks, offering a new perspective on complex and multi-relational systems. </p>