Named entity recognition method in health preserving field based on BERT
Qiang Zhang, Yong Sun, Zhang Li, Yanfei Jiao, Yue Tian
Abstract
With the aging of the population development, people pay more attention to health preserving. In order to build a knowledge graph of health-preserving field, named entity recognition is required first. In the health-preserving field, due to there is no publicly available data sets, building corpus by getting data from websites and defining seven types of entities. Considering the complexity and ambiguity of data, a named entity recognition method based on BERT in the health-preserving field is proposed. Because BERT can generate different vectors for the same word and make full use of context semantic. CNN can extract local features and BILSTM can capture long distance characteristics. Finally selecting the best sentence through the conditional random field. By comparing with other models in the same data set, it is verified that the new model is better in precision, recall and F1 score, reaching the optimal value of 87.84%, which can meet the task requirements in the health-preserving field.