Knowledge Completion Method Based on Relational Embedding with GNN

Knowledge graph, as an effective representation of structured knowledge, plays an increasingly important role in the field of artificial intelligence. However, the incompleteness and sparsity of knowledge graphs have become a bottleneck that restricts their further application. To solve this problem, researchers have proposed some knowledge completion methods based on graph neural network (GNN), which can complete the missing information in the knowledge graph through deep learning techniques. GNN is a kind of neural network approach specialized in processing graph-structured data, which can capture the node features and edge relations in the graph, and then deeply represent the information in the graph. In the knowledge-completion task, the GNN model updates the representation of a node by aggregating its neighborhood information, thus enabling the prediction of missing relations. This paper analyzes some models such as CompGCN and SE-GNN. They have shown excellent performance on several datasets. There is also an analysis of the future development of GNN-based methods for knowledge completion.