融合LDA主题模型和二维卷积的短文本分类  

SHORT TEXT CLASSIFICATION COMBINING LDA TOPIC MODEL AND 2D CONVOLUTION

作　　者：康宸 郑山红[1] 李万龙[1] Kang Chen;Zheng Shanhong;Li Wanlong(Changchun University of Technology,Changchun 130012,Jilin,China)

机构地区：[1]长春工业大学,吉林长春130012

出　　处：《计算机应用与软件》2020年第11期127-131,153,共6页Computer Applications and Software

基　　金：吉林省自然科学基金项目(20130101060JC)。

摘　　要：由于受到短文本文本长度的限制,传统分类模型不能够充分挖掘短文本序列信息,导致短文本分类效果不佳。对此提出两种融合LDA主题模型和二维卷积的短文本分类模型。采用LDA主题模型预训练得到的主题词项分布来弥补短文本缺乏的主题信息,通过预训练词向量来补充短文本通用语义信息。同时将随机初始化词向量、预训练词向量,以及主题词项分布进行拼接,应用二维卷积学习拼接后短文本表示的空间层次结构。实验结果表明,相比于其他短文本分类算法,提出的两种短文本分类模型可以充分挖掘利用短文本特征,其分类准确度明显提升。Due to the limitation of the length of short text,the traditional classification model can not fully exploit the short text sequence information,resulting in poor short text classification.Therefore,this paper proposes two short text classification models combining LDA topic model and 2D convolution.The topic term distribution obtained by pre-training LDA topic model was used to make up for the lack of topic information of short text,and the short text general semantic information was supplemented by pre-training word vector.We spliced the random initialization word vector,the pre-training word vector and the topic term distribution,and applied the 2D convolution learning to construct the spatial hierarchy of the short text representation.The experimental results show that compared with other short text classification algorithms,two short text classification models can fully exploit the short text features,and the accuracy of short text classification was significantly improved.

关 键 词：短文本分类 LDA主题模型 二维卷积 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]