摘要
相关研究数据表明在使用社交网络进行社交活动的95后中有69.8%的网民倾向于使用各类表情符号表述情感倾向。表情符号的高频使用与其自身带有的鲜明情感倾向使得表情符号成为文本情感分析的重要语料资源。基于此,提出了一种多类表情符号的短文本情感模型EMME。模型针对Twitter语料库以5类表情符号融入文本语言进行情感分析,首先利用CBOW模型构建词向量,继而使用卷积对拼接的词向量进行特征融合,后使用MLP实现文本正负情感分类,并针对5类表情符号与文本情感概率进行线性回归。实验数据表明对含各类表情符号短文本情感倾向判别中,EMME模型相比于MNB模型、SVM模型以及EMB模型的MacroF1值分别提高了14.81%、10.42%与9.01%;且EMME模型在不同样本容量规模中均取得了最好的分类准确率。
Relevant research data show that 69.8% of the 95s who use social networks for social activities tend to use various types of emoticons to express their emotional tendencies.The high frequency use of emoticons and their own distinctive emotional tendencies make emoticons an important corpus resource for text sentiment analysis.Based on this,a short text sentiment model EMME with multiple types of emoticons is proposed.The model integrates five types of emoticons into the text language for sentiment analysis in the Twitter corpus.Firstly,the CBOW model was used to construct the word vector,and then the convolution was used to fuse the concatenated word vectors.Then the MLP was used to realize the positive and negative sentiment classification of the text,and the linear regression was performed on the five types of emoticons and the text sentiment probability.The experimental results show that the MacroF1 value of EMME model is 14.81%,10.42% and 9.01% higher than that of MNB model,SVM model and EMB model respectively.The EMME model has achieved good classification results for different sample size.
作者
陈俊
李佳敏
朱丽佳
李丹丹
CHEN Jun;LI Jia-min;ZHU Li-jia;LI Dan-dan(Education College,Guizhou Normal University,Guiyang Guizhou 550025,China;Foreign Languages College,Guizhou Normal University,Guiyang Guizhou 550025,China)
出处
《计算机仿真》
2024年第8期292-295,308,共5页
Computer Simulation
基金
国家自然科学基金(72164004)
贵州师范大学资助博士科研项目([2013]3-21)
贵州省教育厅高校人文社会科学研究项目(2020GH015)。
关键词
情感分析
表情符号
深度学习
词向量
自然语言处理
Emotion analysis
Emoji
Deep learning
Word vector
Natural language processing
