门控空洞卷积与级联网络中文命名实体识别  

Chinese Named Entity Recognition for Gated-dilated Convolution and Cascading Networks

作　　者：谭岩杰 陈玮[1] 尹钟[1] TAN Yan-jie;CHEN Wei;YIN Zhong(School of Ophcal-Electrial and Computer Engineering,Univeristy of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《小型微型计算机系统》2023年第6期1198-1203,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学青年基金项目(61703277)资助.

摘　　要：在中文命名实体识别任务中,文本数据存在多种属性实体,传统模型需要在每个实体上进行繁杂的分类任务,工作量大,难以识别,且采用循环结构,模型运算速度较慢.本文在传统模型的基础上,提出门控空洞卷积与级联网络,门控空洞卷积提高模型运算速度,级联结构将中文命名实体识别任务改为多任务学习,将实体的属性与位置分开标注,降低分类计算量,得到的结果再进行拼接,形成完整的标注结果.实验结果表明,本文提出的门控空洞卷积与级联结构的模型,在Resume数据集上,F1达到了95.50%,比baseline模型提高了1.79%,验证该文提出的模型具有良好的有效性与优越性.In the Chinese named entity recognition task,there are multiple attribute entities in the text data.The traditional model needs to carry out complex classification tasks on each entity,which is difficult to identify due to the heavy workload and the circular structure,and the model operation speed is slow.Based on the traditional model,this paper proposes the gated-dilated convolution and cascade network.The gated-dilated convolution improves the computing speed of the model,and the cascade structure changes the Chinese named entity recognition task to multi-task learning,marks the attributes and positions of entities separately,reduces the amount of classification calculation,and splicing the obtained results to form a complete annotation result.The experimental results show that the model of gated-dilated convolution and cascade structure proposed in this paper can reach 95.50%on Resume data set,which is 1.79%higher than the baseline model.The effectiveness and superiority of the model proposed in this paper are verified.

关 键 词：中文命名实体识别 门控空洞卷积 级联 多任务学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]