Comparative Analysis of Machine Learning Algorithms for Sentiment Classification of Discord App Reviews

The increasing use of digital communication applications such as Discord has generated diverse user opinions expressed through reviews on the Google Play Store. This study aims to analyze user sentiment toward the Discord application using text mining and machine learning techniques. A total of 3,000 reviews were collected through web scraping, pre-processed, labeled using a lexicon-based approach with TextBlob, and balanced using the SMOTE-Tomek method. Sentiment classification was performed into positive, negative, and neutral categories using Decision Tree, Logistic Regression, Support Vector Machine (SVM), and an Ensemble method. The Ensemble model achieved the highest accuracy of 98.67%, followed by Decision Tree (96.50%), SVM (95.83%), and Logistic Regression (90.33%). Limitations of this study include the use of lexicon-based sentiment labeling, machine translation from Indonesian to English, and initial class imbalance. Despite this strong performance, the study has limitations related to lexicon-based labeling, translation of reviews into English, and the presence of a highly imbalanced class distribution in the original dataset. Overall, the findings demonstrate that the Ensemble approach effectively improves sentiment classification accuracy and can support data-driven decision-making in application development.