<![CDATA[The rapid growth of mobile gaming has led to a substantial increase in user-generated reviews, providing valuable insights into user experience. However, many existing sentiment analysis studies on mobile game reviews rely on balanced datasets and pay limited attention to the challenges of multi-class classification under imbalanced conditions, particularly for minority classes such as neutral sentiment. In addition, limited studies systematically examine how class imbalance affects the comparative performance of Machine Learning and Deep Learning models within a unified experimental setting. This study evaluates the performance of Machine Learning and Deep Learning approaches for sentiment analysis using imbalanced mobile game review data. A dataset of 5000 reviews collected from the Google Play Store is categorized into three classes: positive, neutral, and negative. Light Gradient Boosting Machine (LightGBM) and Convolutional Neural Network (CNN) are used as representative models, with class weighting applied to address data imbalance. The findings show that CNN achieves slightly higher accuracy (68.20%) than LightGBM (66.40%), although both models show comparable performance in macro-average metrics. Both approaches experience difficulty in identifying the neutral class, reflecting the impact of class imbalance. These findings emphasize that class distribution plays a more critical role than model choice in real-world sentiment classification.]]>
Copyrights © 2026
