Building and Optimizing Deep Learning Models for Sentiment Analysis in English Text

Sentiment analysis, a key NLP branch, is widely used in social media and market analysis. Traditional methods, reliant on feature engineering and shallow ML models, face limits in text length, diversity, and contextual understanding. Recently, deep learning models, especially those using the Transformer architecture, have become dominant due to their self-attention and parallel computing. This paper introduces a Transformer-based model for English sentiment analysis, studying its construction and optimization. Using datasets like IMDB and Twitter, it achieves efficient classification. Experiments in hyperparameter tuning, regularization, and data augmentation further enhanced performance, with ablation studies assessing each strategy's impact. Results show the model outperforms traditional and mainstream deep learning methods in accuracy, F1 score, etc., especially in long text analysis. This research highlights that optimized deep learning models boost English sentiment analysis performance, offering new insights for related research and applications.