Improving Knowledge Graph Entity Alignment with Graph Augmentation

Entity alignment (EA) which links equivalent entities across different knowledge graphs (KGs) plays a crucial role in fusion. In recent years, graph neural networks (GNNs) have been successfully applied many embedding-based EA methods. However, existing GNN-based methods either suffer from the structural heterogeneity issue that especially appears real KG distributions or ignore heterogeneous representation learning for unseen (unlabeled) entities, would lead model to overfit on few seeds (i.e., training data) and thus cause unsatisfactory performance. To enhance ability, we propose GAEA, novel approach based augmentation. this model, design simple Entity-Relation (ER) Encoder generate latent representations via jointly modeling comprehensive information rich relation semantics. Moreover, use augmentation create two views margin-based contrastive entity learning, mitigating negative influence caused by sparse seeds. Extensive experiments conducted benchmark datasets demonstrate effectiveness of our method. Our codes are available at https://github.com/Xiefeng69/GAEA .