Research on Link Prediction Method Based on Information Fusion Graph Embedding

Research on Link Prediction Method Based on Information Fusion Graph Embedding

To accurately and efficiently capture the topological and attribute information of nodes and apply them to the link prediction task, this paper proposes a Dual Channel Graph Convolution Link Prediction (DC-GCN). DC-GCN constructs a dual channel through the graph convolution network. DC-GCN can learn both topological embeddings and attribute embeddings of nodes; it introduces an attention mechanism to learn the weights of each embedding adaptively and then performs weighted fusion to obtain the final embedding representation of nodes. Finally, the Hadamard distance of nodes is used to construct the link representation between nodes, and the probability of linking between nodes is obtained by training a logistic regression function. By comparing and analyzing many different types of link prediction algorithms, the results show that this algorithm has greater advantages in both AUC and Precision evaluation metrics, so DC-GCN can effectively combine node attributes and structural information of the network to improve the accuracy of the link prediction algorithm.