Leveraging Swin Transformer for advanced sentiment analysis: a new paradigm.

Area of Science:

• Natural Language Processing
• Machine Learning in Healthcare

Background:

• Healthcare text data complexity necessitates advanced sentiment analysis.
• Existing models struggle with local patterns and global dependencies in specialized medical text.

Purpose of the Study:

• To introduce a novel hybrid Swin Transformer-BiLSTM-Spatial MLP (Swin-MLP) model.
• To improve feature extraction and sentiment analysis performance on domain-specific healthcare text.

Main Methods:

• Developed a Swin-MLP model integrating hierarchical attention and shifted-window mechanisms.
• Evaluated the model on Drug Review and Medical Text datasets.
• Compared performance against baseline models including BERT, LSTM, and GRU.

Main Results:

• The Swin-MLP model demonstrated superior performance across accuracy, precision, recall, F1-score, and AUC.
• Achieved a 1-2% improvement in mean accuracy compared to BERT.
• Statistical significance (p < 0.05) confirmed the model's efficacy.

Conclusions:

• The Swin-MLP model is robust and efficient for domain-specific healthcare sentiment analysis.
• Architectural innovations contribute to significant performance gains.
• Future work includes exploring lightweight attention and multimodal analysis.