基于WAE改进模型的文本分类方法研究  

Research on Text Classification Based on WAE Improved Model

作　　者：陈鑫 邱占芝[1] CHEN Xin;QIU Zhan-zhi(school of Mechanical Engineering,Dalian Jiaotong University,Dalian Liaoning 116028,China)

机构地区：[1]大连交通大学机械工程学院,辽宁大连116028

出　　处：《计算机仿真》2022年第10期331-336,共6页Computer Simulation

摘　　要：针对主题模型在文本分类过程中的作用日渐凸显且最早应用于图像领域的WAE模型迁移到自然语言处理领域的应用过程中存在着一些“缺陷”,在WAE模型基础上进行了改进和优化,重点提出了GMWAE和DWAE两种改进模型,并将其与SVM结合起来进行中文的文本分类。在文本分类的研究过程中,进行了是否引入WAE、GMWAE、DWAE主题模型参与文本分类四种场景的搭建。通过实验表明,在SVM分类算法之前增加WAE、GMWAE及DWAE模型时,会在提升分类精度的同时减少分类的时间,其中改进后的模型GMWAE及DWAE表现均优于WAE模型,DWAE表现略佳。In view of the increasingly prominent role of topic model in the process of text classification and some “defects” in the process of moving WAE model, which was first applied in the image field, to the application in the natural language processing field, this paper improves and optimizes the WAE model, and puts forward two improved models, GMWAE and DWAE,and combines them with SVM for Chinese text classification. In the text classification process of this paper, four scenarios whether to introduce WAE,GMWAE,DWAE topic model to participate in the text classification were built.The experiments show that when WAE,GMWAE,DWAE models are added before SVM classification algorithm, the classification accuracy can be improved while the classification time can be reduced. Among them, GMWAE and DWAE models perform better than WAE model and DWAE performs slightly better than GMWAE.

关 键 词：文本分类 主题模型 瓦瑟斯坦自编码器改进模型 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]