Entity alignment with adaptive margin learning knowledge graph embedding

A large number of knowledge graphs have been constructed at present. However, there is diversity and heterogeneity among different knowledge graphs. The relation and attribute of the knowledge graph contain rich semantic information, which helps construct the potential semantic representation of the knowledge graph. At present, the method based on knowledge representation is an important method of entity alignment, which can align entities by transforming them into spatial vectors. And it helps to reduce the heterogeneity among different knowledge domains. However, existing methods use the same optimization goal for triples under different relations, ignoring the difference between relationships. In this article, we put forward a kind of entity alignment method based on the TransE model and use adaptive margin strategies in training. At the same time, this paper studies the LSTM encoder model and the BERT pretraining model in the application of entity alignment. To enhance the model's robustness, we put forward the triple selection strategy based on attribute similarity. Experimental results on real datasets show that this method is significantly improved compared with the baseline model.