一种融合实体类别特征的医疗领域关系抽取方法  

A relation extraction method in the medical field integrating the entity category features

作　　者：游新冬 赵明智 王星予 徐戈 吕学强[1] YOU Xindong;ZHAO Mingzhi;WANG Xingyu;XU Ge;Lü Xueqiang(Beijing Key Laboratory of Internet Culture and Digital Dissemination Research,Beijing Information Science&Technology University,Beijing 100101,China;College of Computer and Control Engineering,Minjiang University,Minjiang 350108,China;Fujian Provincial Key Laboratory of Information Processing and Intelligent Control,Minjiang 350108,China)

机构地区：[1]北京信息科技大学网络文化与数字传播北京市重点实验室,北京100101 [2]闽江学院计算机与控制工程学院,闽江350108 [3]福建省信息处理与智能控制重点实验室,闽江350108

出　　处：《北京信息科技大学学报（自然科学版）》2022年第6期19-25,共7页Journal of Beijing Information Science and Technology University

基　　金：国家自然科学基金资助项目(62171043);国家语委资助项目(ZDI145-10,YB145-3);北京市自然科学基金资助项目(4212020);中央引导地方资助项目(2020L3024)。

摘　　要：关系抽取是自然语言处理中的核心任务,也是构建医疗领域知识图谱中的关键问题。现有的关系抽取方法鲜有融合实体类别的特征,针对医疗领域中实体类别的特点,提出一种融合实体类别特征的医疗领域关系抽取方法CBBS(category BERT BiLSTM Sigmoid)。首先融入实体类别特征,采用基于Transformer的双向编码器表示(bidirectional ecoder representations from Transformers,BERT)-双向长短时记忆(bidirectional long short-term memory,BiLSTM)-Sigmoid模型进行医疗关系抽取,将医疗领域关系抽取问题转化为序列标注问题,提升了单条语料中同一实体处于多种关系时的抽取效果。对比实验表明,CBBS方法在构建的医疗领域关系抽取数据集上与其他方法相比取得了最好的实验效果,精确率达到了83.97%,能够有效地解决医疗领域关系抽取问题。Relation extraction is the core task in natural language processing,and also a key problem in building a knowledge map in the medical field.The existing relation extraction methods rarely integrate the characteristics of entity categories.In view of the characteristics of entity categories in the medical field,a relation extraction method named CBBS(category BERT BiLSTM Sigmoid)in the medical field integrating entity category features was proposed.The CBBS method incorporates entity category features,uses the BERT(bidirectional ecoder representations from Transformers)-BiLSTM(bidirectional long short-term memory)-Sigmoid model for medical relation extraction,and converts the relation extraction problem in the medical field into a sequence labeling problem,which improves the extraction effect when the same entity in a single corpus is in multiple relationships.The comparative experimental analysis shows that the CBBS method has achieved the best experimental results compared with other methods on the constructed medical field relation extraction dataset,with an accuracy rate of 83.97%,which can effectively solve the relation extraction problem in the medical field.

关 键 词：关系抽取 实体类别 BERT 序列标注 混合模型 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]