Analyzing patients satisfaction level for medical services using twitter data.

Area of Science:

• Health Informatics
• Computational Social Science
• Natural Language Processing

Background:

• Public concern for health systems surged due to COVID-19.
• Social media platforms like Twitter are vital for real-time public opinion on healthcare.
• Existing studies using machine learning (ML) and deep learning (DL) for healthcare tweet sentiment analysis show limitations in accuracy and validation.

Purpose of the Study:

• To determine global public sentiment towards medical services using Twitter data.
• To evaluate the effectiveness of various ML and DL models for sentiment classification of healthcare-related tweets.
• To introduce and validate a novel transfer learning-based model for enhanced sentiment analysis.

Main Methods:

• Tweets were collected using hashtags #healthcare, #healthcare services, and #medical facilities.
• Sentiment analysis was performed, considering location and tweet content.
• Multiple ML models (SVM, Logistic Regression, Naive Bayes, ETC, KNN, Random Forest, Decision Tree, AdaBoost) and a proposed LSTM-ETC model were evaluated.

Main Results:

• The Extra Tree Classifier (ETC) model achieved an accuracy of 0.88.
• The proposed transfer learning-based LSTM-ETC model demonstrated superior performance with an accuracy of 0.95.
• Analysis revealed a predominantly positive public sentiment towards healthcare services, with significantly more positive than negative tweets.

Conclusions:

• The proposed LSTM-ETC model offers a highly effective approach for healthcare tweet sentiment classification.
• Public feedback on social media, particularly Twitter, provides valuable insights for improving healthcare quality.
• Positive public sentiment indicates general satisfaction, crucial for strategic decision-making in healthcare.