基于临床疗效评价的中医药论文多标签分类研究  

Research on multi-label classification of Chinese medicine papers based on clinical efficacy evaluation

作　　者：高曼 童元元[1] 刘扬[1] 孙美玲 张雨琪 赵芳华 李彦文[1] 李海燕[1] GAO Man;TONG Yuanyuan;LIU Yang;SUN Meiling;ZHANG Yuqi;ZHAO Fanghua;LI Yanwen;LI Haiyan(Institute of Information on Traditional Chinese Medicine,China Academy of Chinese Medical Sciences,Beijing 100700,China)

机构地区：[1]中国中医科学院中医药信息研究所,北京100700

出　　处：《中国数字医学》2025年第2期14-20,共7页China Digital Medicine

基　　金：中国中医科学院科技创新工程项目(CI2021B002);中国中医科学院基本科研业务费自主选题项目(ZZ160322);中国中医科学院创新工程重大攻关项目(CI2021A00502)。

摘　　要：目的:利用深度学习和多标签分类技术,对论文题录信息实施多学科分类,促进实现中医药领域文献自动分类。方法:选取《2022年度中医医院学科(专科)学术影响力评价研究报告》21万余条中文论文题录信息作为样本,采用BERT、CNN、BiLSTM、CNN-BiLSTM等深度学习方法,构建多标签分类模型进行实验,并对比不同模型和特征组合的实验效果。结果:中医药学术论文的多学科自动分类效果最好的模型为BiLSTM和CNN,最佳特征组合为题名+关键词+摘要,两个模型Macro-F1值分别达95.81%和95.06%。结论:当前深度学习模型多样,受到训练数据、任务特性等方面影响,不同模型在相同任务中表现效果会有所差异,因此实施文本分类时可采用多种不同类型的深度学习模型进行训练对比结果,从而筛选最佳模型与特征组合。Objective To implement multi-disciplinary classification paper title information and promote the automatic classification of literature in the TCM field by use deep learning and multi-label classification technology.Methods More than 210,000 Chinese paper catalogs were selected from the"Research Report on Academic Impact Evaluation of Disciplines(Specialties)in TCM Hospitals in 2022"as samples.BERT,CNN,BiLSTM,CNN-BiLSTM and other deep learning methods were used to establish multi-label classification models for experiments,and the experimental effects of different models and feature combinations were compared.Results BiLSTM and CNN were the best models for automatic classification of TCM academic papers,and the best feature combination was title+keywords+abstract.The Macro-F1 values of the two models were 95.81%and 95.06%,respectively.Conclusion Currently,there are various deep learning models that are affected by training data and task characteristics,and other factors.Different models may perform differently in the same task.Therefore,when implementing text classification,multiple types of deep learning models can be used for training comparison results,so as to screen the best model and feature combination.

关 键 词：中医药 临床疗效评价 多标签分类 文本分类 深度学习 

分 类 号：R197.3[医药卫生—卫生事业管理] R319[医药卫生—公共卫生与预防医学]