Manual annotation based sentiment analysis of user feedback in health and wellness app reviews

Abstract

Abstract This study examines end-user feedback on the Health and Wellness mobile health application, sourced from the play store, by integrating sentiment classification and linguistic intensity analysis to evaluate user perceptions. Comments are categorized into five distinct classes, ranging from highly positive (Class 5) to highly negative (Class 1), using keywords and linguistic patterns. The dataset comprises 20,651 rows, with 9063 highly positive comments, 5877 moderately positive comments, 1380 neutral comments, 298 moderately negative comments, and 4033 highly negative comments. To assess classification efficacy, several state-of-the-art algorithms were implemented, encompassing Support Vector Machines (SVM), Naive Bayes, ensemble-based classifiers (Decision Tree and Random Forest), and deep learning architectures (Convolutional neural networks, CNN). The experimental outcomes underscore the comparative advantage of these approaches in sentiment classification tasks, with CNN achieving the highest accuracy in detecting subtle contextual features. This analysis contributes to application design processes by providing insights into user interaction, satisfaction drivers and communication strategies for digital health platforms.