A Link Prediction Method Based on Graph Neural Network Using Node Importance

Link prediction is a challenging task in complex networks and data mining Its primary purpose is to predict the possibility of links in the future. Link prediction has many application scenarios, such as product recommendations on e-commerce platforms, friend mining on social platforms, etc. Existing link prediction methods focus on utilizing neighbor and path information, ignoring the contribution of link formation of different node importance. For this reason, we propose a novel link prediction method based on node importance. The importance of node is calculated by using the topology structure of the directed network and the path information between nodes, and a graph convolutional network model suitable for directed graphs is designed. The importance of nodes is used to control the model to aggregate the neighbor information, thereby generating the vector representation of the node and obtaining the prediction score through the multi-layer perceptron (MLP). We investigate the proposed method and conduct extensive experiments on 6 real-world networks from various domains. The experiments results illustrate that the proposed method outperforms existing state-of-the-art methods.