基于自然语言处理的中文产科电子病历研究  

The Study of Chinese Obstetric Electronic Medical Records Based on Natural Language Processing

作　　者：张坤丽[1] 马鸿超 赵悦淑[2] 昝红英[1] 庄雷[1] ZHANG Kunli;MA Hongchao;ZHAO Yueshu;ZAN Hongying;ZHUANG Lei(School of Information Engineering, Zhengzhou University, Zhengzhou 450001 , China;The Third Ajfiliated Hospital, Zhengzhou University, Zhengzhou 450052, China)

机构地区：[1]郑州大学信息工程学院,河南郑州450001 [2]郑州大学第三附属医院,河南郑州450052

出　　处：《郑州大学学报（理学版）》2017年第4期40-45,共6页Journal of Zhengzhou University:Natural Science Edition

基　　金：973课题(2014CB340504);国家自然科学基金项目(61402419;60970083);国家社会科学基金项目(14BYY096);河南省科技厅基础研究项目(142300410231;142300410308)

摘　　要：电子病历中蕴含着大量的医疗知识和患者的健康信息,而产科电子病历的结构化及信息抽取对临床决策支持及提高人口的生育健康水平具有重要意义.首先对中文产科电子病历的结构特点及内容进行了分析,并采用基于规则的方法对电子病历数据进行了清洗和结构化;其次采用最大熵(ME)模型及基于规则方法按治疗类型对电子病历进行分类,分类的F值达到88.16%;最后,为了进一步利用电子病历进行信息抽取和知识挖掘,以短句为单位,相似度为衡量标准,采用支持向量机(SVM)模型对首次病程记录进行去重处理及自动差异化分析,从分析的结果中筛选出68.6%的重复及相似短句.tion. The structuralization and information extraction of obstetric electronic medical records is of great sig-nificance on clinical decision and the bearing health. The structural characteristics and content of Chinese obstetric electronic medical records were analyzed. The EMR data was cleaned and structuralized by u-sing the rule-base method. Then the electronic medical records of different treatment types were automati-cally classified by using the maximum entropy model and rule-based methods. And the F value reached 88. 16% . At last, in order to further use electronic medical records for information extraction and knowl-edge mining, the support vector machine model, in which a phrase was taken as a unit and similarity as benchmark, was used to remove the repetition in first course of disease records. And the result was that 68. 6% of the reduplicate and similar phrases were deleted from the records. I t was expected that this study could contribute to the further research on the information extraction from obstetrics electronic medi-cal records.

关 键 词：产科电子病历 数据清洗 分类 差异化 相似度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]