<p>In complex network analysis, link prediction is crucial for understanding the structure and dynamics of networks, as it helps identify potential connections. Existing heterogeneous graph distance encoding models based on shortest path often lose important topological features during the encoding process. To handle this dilemma, we propose a novel distance encoding method based on node sequences, aimed at improving both the accuracy and efficiency of link prediction in heterogeneous graphs. Our method first uses one-hot encoding to represent different node types, encoding the nodes in the order they appear along the shortest path between neighboring nodes and the target nodes, thereby forming the node sequence encoding. We then introduce a custom graph neural network architecture and employ a multi-layer perceptron to perform the link prediction task based on node sequence encoding.To demonstrate the effectiveness of our algorithm, we conducted extensive experiments on 4 public heterogeneous graph datasets, involving the prediction of 5 types of connection relationships. The results show that our model, which relies solely on the graph structure, achieves an average accuracy improvement of 1.3% compared to existing state-of-the-art models in link prediction tasks for heterogeneous networks. This research not only offers a new solution for link prediction in heterogeneous graphs but also provides a fresh perspective for analyzing and utilizing complex network structures.</p>