Graph Neural Networks (GNNs) have proven to be highly effective for various graph-related tasks, including link prediction. However, most existing GNN-based methods are designed for single-layer graphs, which include only nodes and links of a single type. This limitation poses a significant challenge, as many real-world applications, such as social networks, involve multilayer graphs with multiple types of edges between nodes. Addressing the need for effective multilayer link prediction is crucial for better performance and insights in these complex networks. To tackle this issue, we have proposed a novel method for multilayer link prediction using GNNs. Our approach begins with transformation multilayer networks into flat network by leverages three different kinds of features: graph features, community features, and embedding features. By integrating these features with the most effective GNN model, we can capture the intricacies of multilayer graphs. We employ a Multilayer Perceptron (MLP) as the decoding mechanism, which enhances the prediction process. This methodology ensures a comprehensive analysis of the multilayer graph structure, facilitating more accurate link predictions. We evaluated our proposed model on six real-world multilayer datasets, demonstrating its effectiveness in handling the complexities of multilayer link prediction. Our results show that our model outperforms other existing models, highlighting its robustness and reliability. The successful application of our method to these diverse datasets underscores its potential for broad applicability in various real-world scenarios, marking a significant advancement in the field of graph-based machine learning.
