基于扩展特征矩阵和双层卷积神经网络的微博文本情感分类  

WEIBO TEXT SENTIMENT CLASSIFICATION BASED ON EXTENDED FEATURE MATRIX AND DOUBLE-LAYER CONVOLUTION NEURAL NETWORK

作　　者：李卫疆[1] 伊靖 Li Weijiang;Yi Jing(Information Engineering and Automation Institution,Kunming University of Science and Technology,Kunming 650000,Yunnan,China)

机构地区：[1]昆明理工大学信息工程与自动化学院

出　　处：《计算机应用与软件》2019年第12期150-155,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(61363045)

摘　　要：针对现有深度学习方法在中文微博短文本分类任务中存在的数据稀疏、忽略微博文本中的表情和词语特征等问题,提出一种基于扩展特征矩阵和双层卷积神经网络的微博文本情感分类算法Dual CNN。针对微博用户常用的微博表情和多种词语特征,建立扩展特征矩阵;将融合扩展特征矩阵后的词向量,分别使用不同的文本编码方式输入卷积神经网络的两层,得到情感分类结果。通过在COAE2014任务4上的对比实验证明,Dual CNN算法取得了93.35%的分类准确率。相比于单层卷积神经网络算法和SVM等传统机器学习算法,Dual CNN模型具有明显的优势。Existing deep learning methods have some problems in short text categorization task of Chinese micro-blog,such as data sparseness,ignorance of expression and word features in micro-blog text,etc.To solve these problems,this paper proposed a microblog text sentiment classification algorithm based on extended feature matrix and double-layer convolutional neural network.We called it Dual-CNN.An extended feature matrix was established for commonly-used expression and word features of weibo by weibo users.The word vectors fused with the extended feature matrix were input into the two layers of convolutional neural network by using different text encoding methods,and the emotional classification results were obtained.The comparative experiments on COAE2014 task 4 show that Dual-CNN algorithm achieves 93.35%classification accuracy.Compared with traditional machine learning algorithms such as single-layer convolutional neural network algorithms and SVM,Dual-CNN has obvious advantages.

关 键 词：卷积神经网络 中文微博 情感分类 扩展特征矩阵 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]