<![CDATA[As artificial intelligence (AI) technologies become more widely adopted, users increasingly express their experiences through online reviews, providing valuable insights into user satisfaction and opportunities for application improvement. This study compares the performance of Support Vector Machine (SVM), Naïve Bayes (NB), and K-Nearest Neighbor (KNN) for sentiment classification of Dola AI user reviews and applies Latent Dirichlet Allocation (LDA) to identify the dominant topics within each sentiment category. The research followed the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework using 28,226 Google Play Store reviews collected through web scraping. The reviews were represented using Term Frequency–Inverse Document Frequency (TF-IDF), class imbalance was addressed using the Synthetic Minority Oversampling Technique (SMOTE), and model performance was evaluated through 5-fold stratified cross-validation using accuracy, weighted precision, weighted recall, and weighted F1-score. The results showed that SVM achieved the highest average performance, with an accuracy of 77.68% and a weighted F1-score of 80.02%, followed by Naïve Bayes with a weighted F1-score of 78.60% and K-Nearest Neighbor with 70.42%. Furthermore, LDA identified six topics in positive reviews, three topics in neutral reviews, and two topics in negative reviews. Positive reviews mainly highlighted the application's ease of use, AI capabilities, and conversational features, whereas negative reviews primarily focused on application performance, the quality of AI-based image editing, and advertisements. These findings provide deeper insights into user perceptions and offer data-driven recommendations for improving the quality and user experience of the Dola AI application.]]>
Copyrights © 2026
