| 摘 要: 应用卷积神经网络分类文本是自然语言处理领域的研究热点，针对神经网络输入矩阵只提取词粒度层面 的词向量矩阵，忽略了文本粒度层面整体语义特征的表达，导致文本特征表示不充分,影响分类准确度的问题。本文提 出一种结合word2vec和LDA主题模型的文本表示矩阵，结合词义特征和语义特征，输入卷积神经网络进行文本分类， 以丰富池化层特征，达到精确分类的效果。对本文提出模型进行文本分类实验，结果表明，本文算法相比传统特征输入 的卷积神经网络文本分类，在F度量值上取得一定程度的提升。 |
| 关键词: 卷积神经网络；主题模型；LDA；word2vec |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 本文受国家自然科学基金(NO.61702063)，重庆市重大科技项目(cstc2013jcsf-jcssX0020). |
|
| A Study of the Short Text Classification with CNN and LDA |
|
ZHANG Xiaochuan,YU Linfeng,SANG Ruiting,ZHANG Yihao
|
( School of Computer Science and Engineering, Chongqing University of Technology, Chongqing 401320)
|
| Abstract: The application of convolution neural network to classify texts is a research hotspot in the field of natural language processing.The traditional input matrix only extracts the word vector matrix in the word granularity level,neglects the expression of the whole semantic feature of the text granularity level,which leads to the problem of insufficient text features representation.This paper proposes a text representation matrix,which combines word2vec and LDA topic model,not only considers the word meaning and ,but also combines thematic semantic features,and inputs CNN to classify the text,so as to enrich the characteristics of the pool layer and achieve the effect of precise classification.The text classification experiment shows that proposed method achieves a certain degree of improvement in F value compared with KNN and SVM classification algorithms. |
| Keywords: convolution neural network;theme model;LDA;word2vec |
