Comparative Sentiment Analysis of TheoTown Reviews on Steam and Google Play Store Using Support Vector Machine

Abstrak

The rapid expansion of the digital gaming industry has led to a substantial increase in player-generated textual reviews across major distribution platforms such as the Google Play Store and Steam. These reviews offer valuable insights into user experiences and perceived game quality; however, their large volume renders manual analysis inefficient. This study explores cross-platform sentiment patterns of TheoTown using a Support Vector Machine (SVM) classification approach. A dataset comprising 24,754 Google Play Store reviews and 1,452 Steam reviews, collected between January 2021 and December 2025, was processed using a structured pipeline that included text cleaning, normalization, tokenization, stopword removal, and stemming, followed by TF-IDF feature extraction. The findings indicate that a linear SVM model delivers consistently strong performance across both platforms, achieving F1 Scores exceeding 97%. Nevertheless, differences appear in probabilistic evaluation, where the Google Play Store dataset attains a higher AUC (0.8501) than Steam (0.6114). Both datasets are highly dominated by positive sentiment (above 94%), yet the Steam model fails to detect negative instances, highlighting the effects of severe class imbalance and limited data. These results emphasize that platform ecosystems influence both sentiment expression and model performance, underscoring the importance of cross-platform analysis.