基于《伤寒论》的命名实体识别研究

Research on Named Entity Recognition Based on Treatise on Febrile Diseases

研究《伤寒论》中命名实体的识别方法,助力张仲景《伤寒论》不同版本文本的深度挖掘,有助于传承中医文化。该文尝试构建ALBERT-BiLSTM-CRF模型,提取《伤寒论》中疾病、证候、症状、处方、药物等实体,并与BiLSTM-CRF模型和BERT-BiLSTM-CRF模型进行对比。五次实验ALBERT-BiLSTM-CRF模型三个评价指标准确率(P),召回率(R)和F1-测度值(F1-score)的平均值分别为85.37%,86.84%和86.02%,相较于BiLSTM-CRF模型和BERT-BiLSTM-CRF模型F1-score分别提升了6%和3%。实验表明相比BiLSTM-CRF和BERT-BiLSTM-CRF模型,ALBERT-BiLSTM-CRF模型在基于《伤寒论》的实体识别任务中效果最好,更适用于中文古籍的知识挖掘。

By studying the identification methods of named entities in treatises on febrile diseases,it will help the in-depth excavation of different versions of treatises on febrile diseases by Zhang Zhongjing and contribute to the inheritance of traditional Chinese medicine culture.The paper attempts to construct ALBERT-BiLSTM-CRF model,extract the disease,syndrome,symptoms,prescription,drugs and other entities are extracted in treatises on febrile diseases,and it is compared them with BiLSTM-CRF model and BERT-BiLSTM-CRF model.The avevage of precision(P),recall(R)and F1-score of the ALBERT-BiLSTM-CRF model in the five experiments are 85.37%,86.84%and 86.02%,respectively.Compared with the BiLSTM-CRF model and the BERT-BiLSTM-CRF model,the F1-score value increases by 6%and 3%,respectively.The experimental results show that compared with BiLSTM-CRF and BERT-BiLSTM-CRF,the ALBERT-BiLSTM-CRF model has the best effect in the entity recognition task based on treatises on febrile diseases,and is more suitable for the knowledge mining of ancient Chinese books.