基于LDA与深度神经网络的维吾尔文情感分类  

Uighur Sentiment Classification Based on LDA and Deep Neural Network

作　　者：买买提阿依甫 吾守尔·斯拉木[1] 帕丽旦·木合塔尔[1] 杨文忠[1] Maimaitayifu;SILAMU Wushouer;MUHETAER Palidan;YANG Wen-zhong(College of Information Science and Engineering,Xinjiang University,Urumqi Xinjiang 830046,China)

机构地区：[1]新疆大学信息科学与工程学院

出　　处：《计算机仿真》2019年第10期194-201,205,共9页Computer Simulation

基　　金：国家“973”重点基础研究计划基金资助项目(2014CB340506);国家自然科学基金资助项目(61363063);新疆大学多语种重点实验室开放课题(XJDX0905-2013-01)

摘　　要：针对维吾尔论坛文本具有稀疏性、主题不明确性、不规范性等问题,并考虑到普通神经网络只将词粒度层面的词向量作为输入,忽略语义粒度层面的全局语义特征的表示,而导致文档特征表示不充分的现象.提出一种基于主题联合词向量的多通道卷积神经网络的情感分类方法.方法结合word2Vec和LDA模型生成主题特征矩阵,获取语义粒度层面特征信息,以丰富卷积网络的池化层特征,从而提高了情感分类的准确率.在维吾尔文情感二分类和五分类数据集上的实验结果表明,提出的模型相比于传统机器学习方法取得了更好的情感分类性能.Aiming at the problems of sparsity,non-standard,subject ambiguous and the problem that common neural network takes only the word vector at the lexical level as input,and neglects the representation of the global semantic features at the semantic level,which leads to the problem that the document features are not sufficient,this paper presents a sentimental classification method of multi - channel convolution neural network based on topic joint word vector. This method combines the word2 vec and LDA topic model to generate the topic feature matrix,and obtains the global semantic feature information to enrich the max - pooling layer features of the convolutional network,thus improving the accuracy of sentiment classification. The experimental results on the Uyghur emotional two - category and five - category datasets show that the proposed model can achieve better sentimental classification performance than the traditional machine learning method.

关 键 词：情感分类 卷积神经网络 主题概率模型 词向量 维吾尔语 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]