sis 18: e23

Research Article

Hybrid neural network model based on multi-head attention for English text emotion analysis

Download41 downloads
  • @ARTICLE{10.4108/eai.12-11-2021.172103,
        author={Ping Li},
        title={Hybrid neural network model based on multi-head attention for English text emotion analysis},
        journal={EAI Endorsed Transactions on Scalable Information Systems: Online First},
        volume={},
        number={},
        publisher={EAI},
        journal_a={SIS},
        year={2021},
        month={11},
        keywords={hybrid neural network model, multi-head attention, English text emotion analysis},
        doi={10.4108/eai.12-11-2021.172103}
    }
    
  • Ping Li
    Year: 2021
    Hybrid neural network model based on multi-head attention for English text emotion analysis
    SIS
    EAI
    DOI: 10.4108/eai.12-11-2021.172103
Ping Li1,*
  • 1: Department of Public Instruction, Nanyang Medical College, Nanyang City 473000, China
*Contact email: snowycry@qq.com

Abstract

Traditional Convolutional Neural Network (CNN) ignores the contextual semantics information when performing emotion analysis tasks. And CNN will lose a lot of feature information during the maximum pooling operation, which will limit the text classification performance. CNN cannot extract the emotion features of English text more comprehensively, and relies heavily on a large number of language knowledge and emotion resources. In this paper, we propose a hybrid neural network model based on multi-head attention for English text emotion analysis. Firstly, the new model uses multi-head attention to learn the dependence between words and capture the emotion words in the English text. Secondly, the improved bidirectional gated recurrent unit is used to extract different granularity emotion features of English text. According to each emotion category and attention mechanism, feature vectors are generated to construct the emotion feature vector set. Finally, the text emotion categories are judged according to the model attributes. The model is tested on MR, IMDB and SST-5 data sets, the results show that the proposed method has better classification effect compared with other models.