摘要
为了分析突发事件期间网络舆论的情感倾向,以更有效地调节人们的情绪,维护社会稳定。提出了一种融合BERT模型和多通道卷积神经网络的深度学习方法用于细粒度情感分类,以获取更加丰富的文本语义特征信息。通过BERT对输入的文本进行编码,以增强文本的语义特征表示,再通过具有多个不同大小的卷积核的并行卷积层来学习文本特征,捕获文本的深层次特征,提升模型在文本分类的性能。对比实验表明,该模型在准确性、召回率和F 1方面均优于传统的情感分类模型,并能显著改善细粒度情感分类的性能。除此之外,还探究了表情符号对细粒度情感分类模型的影响,实验结果表明表情符号转换成文字后可以增强文本的情感特征提取能力,提升模型分类性能。
In order to analyze the emotional tendency of the network public opinion during the emergency,so as to adjust people s emotions more effectively,and maintain social stability,a deep learning method integrating BERT models and multi-channel convolutional neural networks for fine-grained emotion classification were proposed to obtain more abundant information on text semantic features.The input text was encoded by BERT to enhance the semantic feature representation of the text,and the text features were learned by parallel convolution layers with multiple convolution cores of different sizes to capture the deep features of the text and improve the performance of the model in text classification.Comparative experiments show that the proposed model outperforms the traditional emotion classification model in terms of accuracy,recall rate and F 1 value,and can significantly improve the performance of fine-grained emotion classification.In addition,the influence of emoticons on the fine-grained emotion classification model was also explored.The experimental results show that the conversion of emoticons into text can enhance the emotion feature extraction ability of the text and improve the classification performance of the model.
作者
诸林云
范菁
曲金帅
代婷婷
ZHU Lin-yun;FAN Jing;QU Jin-shuai;DAI Ting-ting(School of Electrical and Information Engineering,Yunnan Minzu University,Kunming 650031,China)
出处
《科学技术与工程》
北大核心
2023年第33期14264-14270,共7页
Science Technology and Engineering
基金
国家社会科学基金(21XSH007)
教育部人文社会科学研究项目(20YJCZH129)。
关键词
网络舆论
情感细粒度分类
BERT模型
多通道卷积神经网络
并行卷积层
network public opinion
emotional fine-grained classification
BERT model
multi-channel convolutional neural network
parallel convolutional layer
