基于fcmpCNN模型的网络文本情感多分类标注  

Multi-category classification for network text sentiment based on fcmpCNN

作　　者：周锦峰 叶施仁[1] 王晖[1] Zhou Jinfeng;Ye Shiren;Wang Hui(School of Information Science ＆ Engineering,Changzhou University,Changzhou Jiangsu 213164,China)

机构地区：[1]常州大学信息科学与工程学院,江苏常州213164

出　　处：《计算机应用研究》2018年第12期3551-3555,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61272367);江苏省科技厅资助项目(BY2015027-12)

摘　　要：针对网络文本情感分析,提出了一种基于全卷积—多池化单元的卷积神经网络模型,实现情感多分类标注。无须手动指定多种上下文窗口大小和尽量保留文本的多层次语义,模型通过堆叠多级全卷积—多池化单元,提取出文本特征向量。该文本特征向量包含多个抽象级别、多种上下文窗口大小和不同层次语义的文本特征。模型最后基于此向量计算情感多分类标注。实验表明,模型的网络文本情感多分类标注正确率达到56. 3%,与同类模型比较,提高了情感多分类标注的正确率。This paper developed a convolutional neural network model,based on full-convolution and multi-pooling,to realize the sentiment classification with multiple labels. Model avoided manually specifying multiple context window sizes and retained more levels of semantic features,while it extracted the text feature vector by stacking a number of units of full-convolution and multi-pooling. The vector contained text features with multiple levels of abstraction,multiple window sizes and different levels of semantics. Finally,the model calculated the multi-label classification based on this vector. The experimental results show that the model achieves classification accuracy of 56. 3%. Compared with the other models,the model improves the accuracy of the sentiment classification with multiple labels.

关 键 词：情感分析 情感多分类标注 卷积神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]