COMPARATIVE STUDY OF MACHINE LEARNING MODELS FOR CLASSIFYING SENTIMENT IN GOOGLE GEMINI APP REVIEWS

The rising number of reviews for the Google Gemini app on the Google Play Store reflects diverse user opinions regarding the performance of this AI-based application. To identify sentiment patterns, this research conducted a comparative study of three classification algorithms—Support Vector Machine (SVM), Naive Bayes, and Random Forest—using 14,479 raw reviews collected through scraping. These reviews then went through several preprocessing steps, including case folding, text cleaning, tokenization, normalization, stopword removal, and stemming. After being labeled based on ratings, the dataset formed a highly imbalanced class distribution, consisting of 11,252 positive reviews and 1,571 negative reviews, and was subsequently split using the Hold-Out method with an 80% training and 20% testing ratio. Evaluation using the Confusion Matrix along with accuracy, precision, recall, and F1-score metrics showed that SVM achieved the best performance, producing 91% accuracy, 93% precision, 97% recall, and a 95% F1-score, outperforming Random Forest and Naïve Bayes, which each reached 90% accuracy. Overall, these results highlight SVM as the most effective algorithm for classifying sentiment in Google Gemini reviews, while the predominance of positive feedback suggests a relatively high level of user satisfaction, although model performance on the minority (negative) class remains a challenge due to data imbalance.