基于BERT-wwm-ext多特征文本表示的经济事件主体抽取方法研究  

Research on Subject Extraction Method of Economic Events based on Multi Feature Text Representation of BERT-wwm-ext

作　　者：陈平[1] 匡尧 陈婧[3] CHEN PING;KUANG YAO;CHEN JING(Wuhan Electric Power Technical College,Wuhan,Hubei 430079;Department of Audit,State Grid Hubei Electric Power co.,LTD,Wuhan,Hubei 430072)

机构地区：[1]武汉电力职业技术学院建设及管理系,湖北武汉430079 [2]国网湖北省电力有限公司审计部内控与信息处,湖北武汉430077 [3]武汉电力职业技术学院,湖北武汉430079

出　　处：《武汉电力职业技术学院学报》2020年第2期45-50,共6页Journal of Wuhan Electric Power Technical College

摘　　要：面向电力行业的经济责任事件主体抽取就是对该领域所涉及的事件的主体进行识别。传统字的向量化缺乏对字语义特征的深层次理解导致主体抽取性能降低,而语言表征的预训练结果会直接影响事件主体抽取的效果。本文提出一种基于中文BERT-wwm-ext嵌入的BIGRU网络进行事件主体抽取。首先,利用中文BERT-wwm-ext将得到的字向量、句子向量和位置向量作为该字最终的向量表示,使字的向量化更加具体。同时采用全词覆盖技术,加强了模型对深层次语言表征的学习能力。然后,将得到的字向量输入到BIGRU网络中,进一步学习上下文语义特征。最后,利用解码函数获得最终的事件主体。本文方法准确度(Accuracy)取得88.29%,结果优于对照组,说明本文提出的模型能有效地提高中文事件主体抽取的准确率。The subject extraction of economic responsibility events for the electric power industry is to identify the subjects of events involved in this field.The lack of a deep understanding of the semantic features of the traditional vectorization of words leads to a decrease in subject extraction performance,and the pre-training of language representations directly affects the effect of event subject extraction.This paper proposes a BIGRU network based on Chinese BERT-wwm-ext embedding for event body extraction.First,the Chinese word of BERT-wwm-ext is used to obtain the word vector,the sentence vector and the position vector as the final vector representation of the word,so that the vectorization of the word is more specific.At the same time,the Whole Word Masking technology is adopted to strengthen the model’s ability to learn deep linguistic representation.Then,the obtained word vector is input into the BIGRU network to further learn the context semantic features.Finally,the final event body is obtained using the decoding function.The accuracy of the method(Accuracy)is 88.29%,better than the result of control group,indicating that the model proposed in this paper can effectively improve the accuracy of Chinese event subject extraction.

关 键 词：BERT BiGRU 事件主体抽取 信息抽取 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]