Link Prediction Algorithm Based On Dual-channel Graph Neural Network

The advancement of network science and technology has propelled humanity into the era of complex networks, and link prediction has emerged as a key focus in complex network research. With the gradual maturity of graph neural network technology, it can automatically learn the representation vectors of nodes based on downstream tasks. Researchers have developed various methods using graph neural networks (GNNs) to address the link prediction task, but these methods have not fully utilized multiple information sources. To tackle the challenges, we design a dual-channel GNN-based link prediction algorithm. This method extracts enclosing subgraphs from two different graphs constructed from adjacency matrix and node feature matrix. Then, it inputs the subgraph information into a Deep Graph Convolutional Neural Network (DGCNN) to learn embeddings of target node pairs. Finally, the embeddings obtained from both channels are merged and fed into a fully connected neural network (FCNN) for prediction. The method has been experimentally tested on multiple networks in this paper. Findings reveal that it outperforms comparative methods and delivers outstanding prediction outcomes.