Development of a patients' satisfaction analysis system using machine learning and lexicon-based methods.

Area of Science:

• Natural Language Processing
• Artificial Intelligence in Healthcare
• Medical Ethics

Background:

• Patient rights are fundamental to medical ethics.
• This study addresses the need for automated analysis of patient feedback to understand satisfaction and identify areas for improvement.
• Sentiment analysis and opinion mining of patient messages are crucial for enhancing healthcare quality.

Purpose of the Study:

• To perform sentiment analysis and opinion mining on patient messages using a hybrid approach.
• To identify positive and negative comments within patient feedback.
• To detect specific hospital wards and staff names mentioned in patient messages.

Main Methods:

• A dataset of 822 Persian patient messages (540 negative, 282 positive) was collected and labeled.
• Pre-processing and feature extraction using Term Frequency-Inverse Document Frequency (TFIDF) and a hybrid Multifeature (MF) + TFIDF approach.
• Evaluation involved six feature selectors, five classifiers, and 5-fold cross-validation, reporting metrics like Accuracy (ACC), Area Under the Curve (AUC), and F1 score.

Main Results:

• The best performance was achieved using Multinomial Naïve Bayes with MF+TFIDF features and SelectFromModel (SFM) feature selection, yielding ACC=0.89, AUC=0.87, and F1=0.92.
• The system successfully identified sentiments (positive/negative) and tagged hospital wards and staff names.
• Comparison between manual evaluation and AI-based assessment showed the effectiveness of the developed system.

Conclusions:

• A combination of lexicon-based methods and machine learning classifiers effectively extracts and categorizes patient sentiments.
• The developed online survey system offers an automated approach to analyze patient satisfaction across different hospital wards.
• This AI-driven method provides an alternative to traditional assessment methods, improving efficiency and objectivity.