A Representation Learning Link Prediction Approach Using Line Graph Neural Networks

Link prediction problem aims to infer the potential future links between two nodes in the network. Most of the existing methods exhibit limited universality and are only effective in specific scenarios, while also neglecting the issue of information loss during model training. To address such issues, we propose a link prediction method based on the line graph neural network (NLG-GNN). Firstly, we employ Node2Vector to learn the latent feature representation vector of each node in the network. Secondly, we extract the local subgraphs surrounding the target link and transform them into the corresponding line graphs. Then, we design a Graph Convolutional Network (GCN) to learn the structural feature representation vector of the node through the line graph. Finally, we combine the latent and structural features through the output layer to predict the target links. We execute extensive experiments on 17 diverse datasets, demonstrating the superior performance and faster convergence of our NLG-GNN method over all baseline methods.