计算机应用研究2024，Vol.41Issue(11)：3295-3302,8.DOI:10.19734/j.issn.1001-3695.2024.04.0104
基于句法依存和注意力机制的多特征多重融合情感分类模型
Multi-feature multiple fusion sentiment classification model based on syntactic dependency and attention mechanism
摘要
Abstract
In response to the difficulty of existing deep learning models in extracting rich semantic information from online re-view,which hinders accurate sentiment extraction,this paper proposed a multi-feature multiple fusion sentiment classification model based on syntactic dependencies and attention mechanisms called MF-SDAM.Firstly,this model utilized syntactic de-pendency relationships to extract the aspect-opinion pairs of information from the text.Then,it employed the dynamic word embedding model BERT to obtain dynamic feature vector representations of online review.Next,based on a dual-channel fea-ture extraction strategy,the model utilized a convolutional neural network(TextCNN)and a bidirectional long short-term memory network with attention mechanism(Att-BiLSTM)to extract local and global semantic features of the text.To further extract global semantic information,it concatenated the text features and the output features of Att-BiLSTM,and used the attention mechanism to weight the sentiment features.Finally,it employed a complementary feature fusion strategy with multi-ple fusion methods to fuse the local and global semantic features,reducing the problem of key information loss.It selected three real publicly available online review datasets from the food delivery and hotel domains for performance verification.Ex-perimental results indicate that the performance of MF-SDAM in sentiment classification tasks for online review is outstanding,with its accuracy and F1-score are superior to the 10 baseline models in most cases.Moreover,it exhibits good robustness for imbalanced datasets.关键词
情感分类/在线评论/多特征多重融合/句法依存/注意力机制Key words
sentiment classification/online review/multi-feature multiple fusion/syntactic dependency/attention mecha-nism分类
信息技术与安全科学引用本文复制引用
夏家莉,余子恺,邓庆山,刘德喜,彭文忠,罗文军..基于句法依存和注意力机制的多特征多重融合情感分类模型[J].计算机应用研究,2024,41(11):3295-3302,8.基金项目
国家自然科学基金资助项目(62272206) （62272206）
江西省教育厅科学技术研究项目(GJJ2200560) （GJJ2200560）
