基于深度学习的《脉经》中医术语命名实体识别研究OA

Objective:Based on deep learning method,this paper studies named entity recognition of terms in an ancient traditional Chinese medicine(TCM)book Pulse Classic.Methods:The book covers a large number of professional terms,the knowledge system is complicated,and the classification of words is difficult.Therefore,we used the combination of transfer learning and BERT to preprocess the Pulse Classic data set,and compared it with Bert-CRF,BiLSTM-CRF and Bert-BilstM-CRF models.Results:The F1 value of named entity recognition of Bert-Bilstm-CRF-radical feature model constructed in this experiment was 84.77%.Compared with BERT-CRF,BiLSTM-CRF and BERT-BiLSTM-CRF models,this model fully considered the professionalism and particularity of the field of Chinese medicine during the construction of word vectors,and learns not only the context,but also the radical features of entity words,with the optimal effect.Conclusion:The Bert-BilstM-CRF-radically feature model can effectively realize the named entity category recognition of terms of TCM ancient books,effectively improve the entity recognition accuracy of Chinese ancient books,lay a technical foundation for the subsequent knowledge map construction,and provide high-quality data support for clinical diagnosis.