Chinese Knowledge Graph Completion based on Improved BERT Model

The existing knowledge graph completion methods ignore semantic information, and there are relatively little researches on Chinese corpus. To solve the above problems, this paper first proposes an improved BERT model based on mean-pooling strategy to effectively solve the problem of insufficient sequence representation ability of BERT model; Then, the entity description information is introduced as the knowledge supplement of the model input, which solves the problem of insufficient background knowledge of the previous model; Finally, the knowledge completion problem is transformed into a classification problem. The experimental results show that this method improves the mean rank (MR) index by 5.83 compared with the traditional method, and the top 10 hit rate (Hit@10) The index is improved by 1.67%, which verifies the effectiveness of the model in Chinese corpus.