BERT-SABT: A Text Sentiment Analysis Model Based on Fusion Self-Attention Mechanism and Dual-Channel Feature Extraction Network

To solve the problem of incomplete feature extraction in the current text sentiment analysis model, combined with the characteristics of Weibo comment text, a BERT-SABT model was proposed to improve the accuracy of Weibo comment sentiment classification. The model uses the BERT pre-training model to dynamically encode the text sequence after data preprocessing. Next, it uses the self-attention mechanism to avoid the loss of important information and the influence of redundant information and performs the overall context and local deep feature extraction through the dual channels combining BiLSTM and TextCNN, respectively. Then, it effectively fuses the features of the two channels through feature fusion operation. Finally, the softmax function is used to output the classification results. Compared with the best performing models, the accuracy of the proposed model was improved by 0.46% and 0.70^%, and the Fl value was increased by 0.57% and 0.72%, on the weibo_senti_lOOk and SMP2020-EWECT datasets, respectively. Results show that BERT-SABT model is superior to the existing excellent sentiment analysis models.