Existing link prediction methods for maritime shipping networks primarily rely on single-dimensional network structural information and fail to comprehensively account for multiple influencing factors. To address this limitation, a coupled similarity index integrating multi-level information is proposed. Specifically, to incorporate the effects of practical node attributes such as port operations, hinterland economic conditions, and maritime distances on link prediction, similarity indices based on node realistic attributes are constructed. The predictive performance of 22 individual similarity indices from four perspectives—local network structure, global network structure, quasi-local network structure, and node realistic attributes—is evaluated using the Area Under the Curve (AUC) metric. The index with the highest AUC value from each perspective is then selected. Subsequently, weighted averaging and grid search are employed to determine the optimal weight combination, thereby constructing the coupled similarity index with the highest predictive accuracy. Experimental results on the global container shipping network demonstrate that the Resource Allocation (RA), Random Walk with Restart (RWR), Superposed Random Walk (SRW), and Similarity based on Port Operations (SPO) indices achieve the best prediction performance within their respective perspectives. The proposed coupled similarity index significantly outperforms all individual indices. In particular, the relatively high weight assigned to the SPO index indicates that node realistic attributes play a dominant role in link prediction. Furthermore, comparison with newly added shipping routes in 2024 shows that the proposed method achieves a prediction accuracy of 45.48%, and the spatial distribution of the predicted routes is highly consistent with the global shipping network pattern, thereby validating the effectiveness of the proposed coupled similarity index.