Comparison of CNN and SVM Algorithms in Sentiment Analysis of Roblox Game Based on Bug, Loading Asset, and Connection Stability Aspects

Abstract

Roblox is one of the most popular mobile gaming platforms; however, its rapid growth has led to an increasing number of user complaints related to technical stability, which significantly affects user experience. Common issues such as bugs, asset loading problems, and unstable connections are frequently expressed in user reviews, making sentiment analysis a valuable approach for identifying critical technical problems in mobile games. This study aims to analyze user sentiment toward these technical aspects and to compare the performance of Support Vector Machine (SVM) and Convolutional Neural Network (CNN) in aspect-based sentiment classification. A total of 12,809 Indonesian-language reviews were collected from the Google Play Store during October 2025. The research methodology included data scraping, text preprocessing (cleansing, tokenization, normalization, stopword removal, and stemming), lexicon-based sentiment labeling, and data balancing using the Synthetic Minority Over-sampling Technique (SMOTE). TF-IDF was used for feature extraction in the SVM model, while word embeddings were applied for the CNN model. The results show that the Bug aspect is the most dominant issue (63.91%), followed by Connection Stability (34.41%) and Asset Loading (1.68%). In terms of classification performance, SVM outperformed CNN, achieving 96% accuracy, precision, recall, and F1-score, whereas CNN obtained an accuracy of 80.63% with an F1-score of 0.81. These findings indicate that SVM combined with TF-IDF features is more effective than CNN for classifying short and informal mobile game reviews and provides useful insights for developers in prioritizing technical improvements.