An Entity Prediction Method for Chinese Medical Knowledge Graph via Bert Sentence Embedding and Classification
Mingxia Gao, Hao Li, Furong Chen · 2023
Automatic Knowledge Graph Completion is becoming the main research direction of Knowledge graph construction. Among, the entity prediction method can complete the complement of entities in RDF triples, and is widely used in the generation process of the Knowledge graph. In order to complete the Chinese medical Knowledge graph, this paper proposes an entity prediction method based on BRRT sentence embedding and classification. This method needs three steps, the first step is to introduce a large-scale medical corpus to fine tune the basic BERT model into a BERT Domain model. The second step is obtaining the sentence embedding through the model for candidate triples. The third step is to obtain the top N candidate entity lists according to the ranking of classifier probabilities of all candidate. In order to verify the effectiveness of this method, a series of experiments are conducted on the BIOS. The experimental results show that the optimal accuracy of the entity prediction method in this paper is 20.5%, which is 7.2% higher than that using Word embedding+distance.