基于预训练语言模型的中文地址命名实体识别  

Chinese Address Named Entity Recognition Based on Pre-training Model

作　　者：李廷元[1] 杨勇 Li Tingyuan;Yang Yong(School of computer science,Civil Aviation Flight University of China,Guanghan 618307)

机构地区：[1]中国民用航空飞行学院计算机学院,广汉618307

出　　处：《现代计算机》2022年第15期81-84,120,共5页Modern Computer

摘　　要：随着深度学习的发展,基于深度学习的命名实体识别抽取过程中,作为基础步骤的预训练模型也愈发受到重视。中文预训练语言模型能够更好地结合语义语境,更加充分地考虑到一词多义的情况,因此该语言模型目前也普遍应用于命名实体识别任务。文中首先介绍了BERT、ERNIE、NEZHA三种预训练模型,之后构建预训练模型、BiGRU及CRF的算法模型,在阿里中文地址要素解析比赛数据集上进行中文地址命名实体识别任务。实验结果表明,NEZHA取得当前预训练语言模型最优的识别结果。Compared with traditional machine learning methods,deep learning methods can achieve better performance in named entity recognition tasks without relying on artificial features.Named entity extraction process based on deep learning,as a basic step of the pre-training model is also more and more attention.The Chinese pre-training model is widely used in named entity recognition,which can better combine semantic context and fully consider the polysemy problem.In this paper,BERT,ERNIE and NEZHA are introduced,the pre-training model,BiGRU and CRF algorithm model structure are built,and the Chinese address named entity recognition task is performed on the dataset of Ali Chinese Address Elements Parsing Contest.Experimental results show that ERNIE obtains the optimal recognition result of the current pre-training model.

关 键 词：预训练模型 命名实体识别 深度学习 中文地址 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]