Application of natural language processing based on Transformer model in sentiment analysis of American Initiation Stories

This paper proposes an innovative algorithm based on multi-layer sentiment attention mechanism to improve the sentiment analysis performance of Initiation Stories works. The multi-layer sentiment attention mechanism (MSA-Transformer) achieves accurate modeling of text sentiment information through three key mechanisms: multi-head sentiment vector extraction, sentiment enhancement module, and context-adaptive sentiment weight allocation. The experiment uses classic American Initiation Stories works such as The Catcher in the Rye, To Kill a Mockingbird, and Wuthering Heights to construct a sentiment dataset containing more than 100,000 labeled sentences. In the experiment, the accuracy of MSA-Transformer in sentiment classification tasks reached 91.3%, which is significantly improved compared with BERT (87.6%), GPT (86.9%), and BiLSTM (82.4%). In addition, the model in this paper is also superior to existing mainstream methods in indicators such as F1 value and recall rate, and shows better stability in long text processing capabilities. The experimental results show that MSA-Transformer has a strong advantage in the sentiment analysis task of Initiation Stories, providing new ideas for the application of natural language processing in literary research.