Optimizing Patient Feedback with Generative Adversarial Network Leveraging Knowledge Distillation to Improve...

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Natural Language Processing

Background:

• Limited utilization of domestic healthcare facilities necessitates patient-centered improvements.
• Patient feedback is crucial for enhancing healthcare accountability, transparency, and quality.
• A systematic approach to patient feedback is needed to inform healthcare policies.

Purpose of the Study:

• To develop a system for collecting and analyzing patient feedback to improve local healthcare.
• To create a platform for reviewing hospital feedback, especially in high-demand medical service areas.
• To establish the "Dhaka Private Hospitals Review Dataset" for methodical patient opinion evaluation.

Main Methods:

• Transformer-based generative adversarial learning for sentiment analysis.
• Knowledge distillation (KD) to enhance model efficiency.
• Proposed GANBERT architecture with GC-BERT and LC-BERT student models, utilizing static pretrained word embeddings.

Main Results:

• GC-BERT improved execution time by 1.27%–24.27%; LC-BERT improved by 14.13%–23.30%.
• Both models demonstrated significant parameter reduction (82.50%–99.99%), offering lightweight and efficient solutions.
• The approach using static pretrained word embeddings proved effective for patient feedback analysis.

Conclusions:

• The developed system and GANBERT architecture offer an efficient method for analyzing patient feedback.
• This approach can significantly enhance healthcare policy and practice by incorporating patient-centered input.
• The "Dhaka Private Hospitals Review Dataset" and proposed models provide a valuable resource for healthcare quality improvement.