Hierarchical Feature Fusion Graph Neural Network Model for Link Prediction

In the field of link prediction in graph neural networks, existing graph neural network models have difficulty effectively integrating node structural features and latent features, which affects the prediction accuracy. When dealing with complex graph structures, traditional methods struggle to balance the expressive ability and computational efficiency. To address these issues, this paper proposes a Hierarchical Feature Fusion Link Prediction model (HFF-LP). The model extracts subgraphs from the network, and after node labeling and line graph transformation, it uses the Graph Attention Network (GAT) to obtain the structural features of the subgraphs and the traditional Graph Convolutional Network (GCN) to obtain the latent features of the entire graph. For the nodes to be predicted, the two types of features are concatenated, and then the node pairs to be predicted are processed by a Multi-Layer Perceptron (MLP). Experimental results show that the HHF-LP can effectively integrate features, improve the prediction accuracy, balance the expressive ability and computational efficiency, and provide new ideas and methods for the research on link prediction in graph neural networks.