摘要
目的:探索分析BERT-BiLSTM-CRF模型抽取中医文献摘要中的实体的可行性及识别效果。方法:在知网数据中导出500条中医疗法治疗功能性胃肠病的论文摘要,对文本中的西医病名、临床表现、方剂、中药等11类实体进行BIO标注,基于BERT-BiLSTM-CRF模型进行训练及参数调整,而后对模型进行测试,并应用于实体识别。结果:模型测试的精确率为85.07%,召回率为88.48%,F1值为0.8674,中药、方剂、西医诊断等实体类别的识别效果较好;模型应用中,自动化实体抽取结果整体较好,能够反映该领域文献的主要研究方向。结论:BERT-BiLSTM-CRF模型能够识别出论文摘要中大部分的实体,可以为知识图谱的自动化构建提供基础,同时也对中医药领域的自然语言处理应用提供了参考和借鉴。
Objective To explore and analyze the feasibility and recognition effect of the BERT-BiLSTM-CRF model in extracting entities from TCM literature abstracts.Methods 500 abstracts of papers on TCM treatment of functional gastrointestinal diseases were extracted from the data of CNKI,and 11 categories of entities in the texts,such as Western medicine disease names,clinical manifestations,prescriptions,and TCM,were BIO-labelled,and the training and parameter adjustment were performed based on the BERT-BiLSTM-CRF model,and then the model was tested and applied to entity recognition.Results The accuracy of model test was 85.07%,the recall rate was 88.48%,and the F1 value was 0.8674,and the recognition effect of entity categories such as TCM,prescription,and western medicine diagnosis was good.In the application of the model,the results of automated entity extraction were overall good,which can reflect the main research direction of the literature in this field.Conclusion The BERT-BiLSTM-CRF model can identify most of the entities in the paper abstracts,which can provide a basis for the automated construction of knowledge graph,and also provide a reference for the application of natural language processing in the field of TCM.
作者
石文艳
赵芳华
孙美玲
李海燕
李敬华
于彤
孔静静
宋源
于琦
SHI Wenyan;ZHAO Fanghua;SUN Meiling;LI Haiyan;LI Jinghua;YU Tong;KONG Jingjing;SONG Yuan;YU Qi(Institute of Traditional Chinese Medicine Information,China Academy of Chinese Medical Sciences,Beijing 100007,China)
出处
《中国数字医学》
2024年第5期78-83,共6页
China Digital Medicine
基金
中国中医科学院科技创新工程项目-名医传承智能化信息平台建设与应用(CI2021A05310)
中国中医科学院自主选题研究项目-中医药知识图谱自动构建共性技术研究与平台开发(ZZ150313)
中国中医科学院科技创新工程项目-基于海量文献的中医脾胃病知识图谱自动构建与知识服务研究(CI2021A05308)。
