Patient Experience Feedback Sentiment Analysis: Combining BERT and LSTM with Genetic Algorithm Optimization

Patient experience is a fundamental component in the evaluation of the provided care's quality. This experience is nearly solely impacted by the interactions the patient has with the environment and medical staff members that they encountered during their hospital stay. Surveys addressed to patients contain a free-text section that allows them to share these experiences in a self-reported format. This results in a significantly large textual data collection. In this research, patient experience feedback collected from their responses to surveys received from Cooper University Hospital following their stay during the entire year of 2023 is investigated. The survey considered is the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey. This paper focuses on conducting the sentiment analysis of these reviews. As a result, the overall sentiment of the hospital stay from the patient's angle is extracted. This allows the hospital to measure their satisfaction level and take a step towards more patient-centered care. Besides, negative comments enable the identification of improvement areas. Before text classification, patient comments are vectorized using BERT to leverage its context-aware representations. Following vectorization, three models, Linear-SVM, MLP, and LSTM, are fitted for polarity classification. The sentiments considered are Positive and Negative which result in a binary classification. The hyperparameters of the Neural Networks-based architectures are afterward tuned using Genetic Algorithm. The models are thereby contrasted using metrics such as F1-score and AUC score. Therefore, sentiments are assigned to patient experience feedback and an understanding from their perspective of the patient's journey is concluded.