Text Classification of Duolingo Reviews on Google Play: Insights for Enhancing M-Learning Applications

Abstract

As digital education tools gained prominence, user feedback played a crucial role in refining and personalizing learning experiences. This study analyzed over 100,000 Google Play reviews of the Duolingo language-learning app, using text classification techniques to extract key insights into user sentiment and preferences. By employing natural language processing (NLP) methods, specifically logistic regression and Naive Bayes classifiers, the study categorized feedback into four primary themes: content, instruction, performance, and user interface and user experience (UI/UX). Logistic regression achieved an AUC score of 0.812, precision of 0.904, recall of 0.900, and F1-score of 0.894, while Naive Bayes achieved an AUC score of 0.806, precision of 0.904, recall of 0.900, and F1-score of 0.894. Both models demonstrated an accuracy rate of 90%. The results indicated that content was the most significant concern for users, comprising 74.2% of all reviews, followed by instructional feedback (14%) and performance issues (9.1%). This analysis provided valuable insights for developers aiming to enhance Duolingo’s user experience by addressing content quality, improving pedagogical approaches, resolving technical issues, and refining the user interface. The findings also contributed to the broader field of educational technology by demonstrating the application of machine learning techniques in understanding user feedback at scale.