融合注意力机制的多通道CNNs-BiLSTM情感极性分析方法  

Multi-channel CNNs-BiLSTM Emotional Polarity Analysis Method Fused with Attention Mechanism

作　　者：谢玉惠 肖桂荣[2] XIE Yu-hui;XIAO Gui-rong(Academy of Digital China(Fujian),Fuzhou University,Fuzhou 350108,China;Key Laboratory of Spatial Data Mining and Information Sharing of Ministry of Education,Fuzhou University,Fuzhou 350108,China)

机构地区：[1]福州大学数字中国研究院(福建),福州350108 [2]福州大学空间数据挖掘与信息共享教育部重点实验室,福州350108

出　　处：《小型微型计算机系统》2023年第6期1140-1145,共6页Journal of Chinese Computer Systems

基　　金：中央引导地方科技发展专项项目(2020L3005)资助;中国科学院战略性先导科技专项子课题项目(XDA231005040)资助.

摘　　要：突发公共卫生事件极易引起社会恐慌,新冠肺炎更是全球聚焦的重大热点事件,客观了解疫情期间的公众情绪响应,有利于政府及相关部门合理管控舆情.本研究以疫情流行期间网民微博博文为基础,通过文本挖掘的方式探索疫情期间网民情感倾向,提出一种以卷积神经网络和双向长短期记忆网络为基础,并融合注意力机制的多通道情感极性分析方法.该方法首先对微博文本数据进行分词和停用词的预处理,通过Word2Vec模型获取词向量表达式,使用多通道CNNs-BiLSTM模型抽取多尺度文本特征,融合注意力机制调整特征权重,以语义相关度进行文本情感倾向判断.通过COVID-19微博舆情数据开展实验验证,结果表明,该方法相较于其他基准模型获得了较高的准确率,能够充分利用多维矩阵捕获丰富的文本特征,具有一定的优越性.Public health emergencies can easily cause social panic.COVID-19 is a major hot event globally.An objective understanding of public emotional response during the epidemic is conducive to the reasonable management and control of public opinion by the government and relevant departments.This study is based on the Weibo of netizens during the COVID-19 epidemic period,and explores the emotional tendencies of netizens during the epidemic through text mining.Propose a multi-channel emotional polarity analysis method based on convolutional neural network and bidirectional long short-term memory network,and fusion attention mechanism.This method first performs word segmentation and stop word preprocessing on the microblog text data,obtains the word vector expression through the Word2Vec model,uses the multi-channel CNNs-BiLSTM model to extract multi-scale text features,and then integrates the attention mechanism to adjust the feature weights.Semantic relevance is used to judge the sentiment tendency of the text.Experimental verification was carried out through COVID-19 microblog public opinion data.The results show that this method has a higher accuracy rate than other benchmark models,and it can make full use of the multi-dimensional matrix to capture rich text features,and has certain advantages.

关 键 词：情感极性分析 双向长短期记忆网络 注意力机制 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]