基于卷积神经网络的互联网短文本分类方法  

Internet Short-text Classification Method Based on CNNs

作　　者：郭东亮[1] 刘小明[1] 郑秋生[1] 

机构地区：[1]中原工学院计算机学院,河南郑州450007

出　　处：《计算机与现代化》2017年第4期78-81,共4页Computer and Modernization

基　　金：河南省科技攻关项目(132102310284);河南省教育厅科学技术研究重点项目(14A520015)

摘　　要：互联网短文本的分类是自然语言处理的一个研究热点。本文提出一种基于卷积神经网络(Convolutional Neural Networks,CNNs)互联网短文本分类方法。首先通过Word2vec的Skip-gram模型获得短文特征,接着送入CNNs中进一步提取高层次特征,最后通过K-max池化操作后放入Softmax分类器得出分类模型。在实验中,该方法和机器学习方法以及DBN方法相比,结果表明本文方法不仅解决了文本向量的维数灾难和局部最优解问题,而且有效地提高了互联网短文本两级分类准确率,证实了基于CNNs的互联网短文本分类的有效性。The Internet short-text classification is a hot research topic in natural language processing. This paper presents a short text classification method based on deep learning＇s convolutional neural networks. First short-text features are achieved by the Skip-gram model of Word2vec,then it is sent into the CNNs to extract high-level features,after the K-max pooling,it is put into the Softmax classifier to get a classification model. In the Internet short-text classification experiments,compared to machine learning and DBN＇s method,the results show that the proposed method not only solves the problems of the curse of dimensionality of text vector and the local optimal solution,but also effectively improves the accuracy of Internet short-text classification,and confirms the validity of the Internet short-text classification method based on CNNs.

关 键 词：卷积神经网络 短文本分类 深度学习 机器学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]