LDA特征扩展的多类SVM短文本分类方法研究  

Research on the Classification Methods of Multiple SVM Short Texts based on LDA Feature Extension

作　　者：郑腾 吴雨川 ZHENG Teng;WU Yu-chuan(School of Mechanical Engineering and Automation,Wuhan Textile University,Wuhan Hubei 430200,China)

机构地区：[1]武汉纺织大学机械工程与自动化学院,湖北武汉430200

出　　处：《武汉纺织大学学报》2019年第2期72-76,共5页Journal of Wuhan Textile University

基　　金：国家自然科学基金面上项目(61271008)

摘　　要：针对短文本信息量少、特征稀疏的特点,提出一种基于LDA主题扩展的多类SVM短文本分类方法。在短文本基础上,利用LDA主题模得到文档的主题分布,将主题中的词扩充到原短文本的特征中,在特征空间上使用基于经典权重计算方法的多类SVM分类器进行分类。实验结果表明,在各个类别上的查准率、查全率和F1值都有所提高,验证了该方法的可行性。Based on the short text and characteristics of sparse, this paper puts forward a short text classify method based on characteristics extend of LDA. This method used the LDA model to obtain the subject distribution of document, extended the word under the corresponding topic into the characteristics of the original short text as a new part of the feature word. A multi-class SVM classifier based on classical weight calculation was used. Experimental results show that the precision, recall and F1 values are improved in all categories. It verifies this model has some superiority in text categorization.

关 键 词：短文本分类 特征扩展 SVM LDA 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]