面向互联网问诊文本的医疗事件时序关系抽取  

TEMPORAL RELATIONSHIP EXTRACTION OF MEDICAL EVENTS ORIENTED TO INTERNET CONSULTATION TEXT

作　　者：凌佳君 刘宇[1,2,3] 顾进广 Ling Jiajun;Liu Yu;Gu Jinguang(Department of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065,Hubei,China;Big Data Science and Engineering Research Institute,Wuhan University of Science and Technology,Wuhan 430065,Hubei,China;Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial,Wuhan 430065,Hubei,China)

机构地区：[1]武汉科技大学计算机科学与技术学院,湖北武汉430065 [2]武汉科技大学大数据科学与工程研究院,湖北武汉430065 [3]湖北省智能信息处理与实时工业系统重点实验室,湖北武汉430065

出　　处：《计算机应用与软件》2023年第11期186-193,240,共9页Computer Applications and Software

基　　金：国家自然科学基金项目(U1836118,61673304);国家社科基金重大计划项目(11&ZD189);湖北省自然科学基金项目(2018CFB194)。

摘　　要：互联网问诊文本中存在大量的医疗事件及其时序关系,这些信息是推断患者患病诊疗时间线的关键信息,但问诊文本通常有不规范的特点。针对不规范的问诊数据,在注意力双向LSTM模型的基础上,添加关系发现词特征为关系识别提供了重要时序线索,弥补了不规范数据中针对两个医疗事件发生时间的描述缺失;添加时序背景知识对模型预测的错误结果进行关系修正,提升了关系抽取效果。从好大夫网站获取到患者真实的问诊文本作为数据进行实验。实验表明,该方法在各种时序关系抽取的F1值上都优于现有方法,提升效果在4%~16%之间。There are a large number of medical events and their temporal relationships in the consultation text generated by the Internet consultation.This information is the key information for inferring the patient's diagnosis and treatment timeline,but the consultation text usually has irregular characteristics.Regarding the irregular consultation data,on the basis of the attention bidirectional LSTM model,the relationship discovery word feature was added to provide important temporal clues for relationship recognition,and to make up for the lack of description of the occurrence time of two medical events in the irregular data.Temporal background knowledge was added to correct the relationship between incorrect results of model prediction and improve the effect of relationship extraction.The real patient's medical consultation text was obtained from the Haodaifu website as data for experiment.Experiments show that this method is superior to the existing methods in the F1 value extracted from each temporal relationship,and the improvement effect is between 4%and 16%.

关 键 词：关系抽取 医疗事件 时序关系 关系发现词 时序背景知识 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]