基于双通道特征融合的微博情感分析  

Sentiment analysis of microblogs based on dual-channel feature fusion

作　　者：胥桂仙[1] 王家诚 张廷[1] 田媛 XU Gui-xian;WANG Jia-cheng;ZHANG Ting;TIAN Yuan(Key Laboratory of Ethnic Language Intelligent Analysis and Security Governance of MOE,School of Information Engineering,Minzu University of China,Beijing 100081,China)

机构地区：[1]中央民族大学信息工程学院,民族语言智能分析与安全治理教育部重点实验室,北京100081

出　　处：《东北师大学报（自然科学版）》2024年第4期62-71,共10页Journal of Northeast Normal University(Natural Science Edition)

基　　金：国家社会科学基金资助项目(19BGL241);国家自然科学基金资助项目(62236011);北京市社会科学基金资助项目(20YYB011)。

摘　　要：提出一种基于双通道特征融合的微博情感分析模型.首先将通过BERT预训练语言模型获取的动态词向量作为情感分类模型的输入;然后使用双通道特征提取网络进行特征提取,一方面使用TextCNN-Attention提取文本局部特征,另一方面使用基于图卷积神经网络的神经主题模型提取文本全局主题特征;接着将局部特征和全局特征融合得到最终的文本向量;最后通过Softmax输出情感极性.在构建的微博评论文本数据集上进行实验,本文模型F1值达到91.36%,相比主流基线模型提升0.73%~8.82%,验证了本文模型在情感分析任务上的有效性.This paper proposes a microblog sentiment analysis model based on dual-channel feature fusion.Firstly,the dynamic word vectors obtained from the BERT pre-trained language model are used as the input of the sentiment classification model,and then a dual-channel feature extraction network is used for feature extraction,which extracts the local features of the text using TextCNN-Attention on one hand,and extracts the global thematic features of the text using the neural thematic model based on graph convolutional neural network on the other hand.Then the local and global features are fused to get the final text vector,and finally the sentiment polarity is output by Softmax.Experiments are conducted on the constructed microblog comment text dataset,and the F1 value of the model reaches 91.36%,having an improvement of 0.73%~8.82%compared with the mainstream baseline models,which verifies the effectiveness of the model on the task of sentiment analysis.

关 键 词：情感分析 预训练语言模型 图卷积神经网络 神经主题模型 特征融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]