基于混合神经网络的中文隐式情感分析
Implicit Sentiment Analysis for Chinese Texts Based on a Hybrid Neural Network
作者:赵容梅(成都信息工程大学网络空间安全学院, 成都 610225);熊熙(成都信息工程大学网络空间安全学院, 成都 610225);琚生根(四川大学计算机学院);李中志(成都信息工程大学网络空间安全学院, 成都 610225);谢川(成都信息工程大学网络空间安全学院, 成都 610225)
Author:ZHAO RongMei(College of Cybersecurity, Chengdu University of Information Technology, Chengdu 610225, China);XIONG Xi(College of Cybersecurity, Chengdu University of Information Technology, Chengdu 610225, China);jushenggen(Sichuan university School of computer science,);LI ZhongZhi(College of Cybersecurity, Chengdu University of Information Technology, Chengdu 610225, China);XIE Chuan(College of Cybersecurity, Chengdu University of Information Technology, Chengdu 610225, China)
收稿日期:2019-09-28 年卷(期)页码:2020,57(2):264-270
期刊名称:四川大学学报: 自然科学版
Journal Name:Journal of Sichuan University (Natural Science Edition)
关键字:情感分析;深度学习;卷积神经网络;注意力机制;长短期记忆网络
Key words:Emotional analysis; Deep learning; Convolutional neural network; Attention mechanism;Long term and short term memory network
基金项目:国家自然科学基金(81901389); 中国博士后科学基金(2019M653400); 四川省科技计划项目(2018GZ0253, 2019YFS0236, 2018GZ0182, 2018GZ0093, 2018GZDZX0039)
中文摘要
隐式情感分析是情感计算的重要组成部分,尤其是基于深度学习的情感分析近年来成为了研究热点.本文利用卷积神经网络对文本进行特征提取,结合长短期记忆网络(LSTM)结构提取上下文信息,并且在网络中加入注意力机制,构建一种新型混合神经网络模型,实现对文本隐式情感的分析.混合神经网络模型分别从单词级和句子级的层次结构中提取更有意义的句子语义和结构等隐藏特征,通过注意力机制关注情绪贡献率较大的特征.该模型在公开的隐式情感数据集上分类准确率达到了77%.
英文摘要
Implicit sentiment analysis is an important part of emotional computing, especially sentiment analysis based on deep learning has become a research hotspot in recent years. This paper uses convolutional neural network to extract features from text, combines long short term memory network (LSTM) structure to extract context information, and adds attention mechanism to the network to construct a new hybrid neural network model to realize implicit emotions analysis on text. The hybrid neural network model extracts more meaningful semantic features such as sentence semantics and structure from the hierarchical structure of word level and sentence level respectively, attention is paid to the characteristics of large emotional contribution rate through attention mechanism. The proposed model has a classification accuracy of 77% on the public implicit sentiment data set and can improve the effect of text emotion analysis more comprehensively.
【关闭】