Sentiment Analysis on Game Review on the Steam Platform using Support Vector Machine, TF-IDF and Chi-Square Methods
Gandhi Risyad Abimanyu, Mahendra Dwifebri Purbolaksono, Adiwijaya Adiwijaya · 2025
As online platforms like Steam grow, user reviews have become a valuable resource for both players and developers. This study presents a sentiment analysis approach combining Support Vector Machine (SVM) for classification, Term Frequency–Inverse Document Frequency (TF-IDF) for feature extraction, and Chi-Square for feature selection. The dataset used in this study consists of Steam game reviews labeled according to user vote polarity, categorized into positive and negative sentiments. Several preprocessing techniques were evaluated, including lemmatization, stopword removal, and slang normalization, to determine their impact on model performance. Twelve experimental combinations were tested by varying pre-processing methods, feature selection, and SVM kernels (RBF, polynomial, and linear). The results showed that non-lemmatized text slightly outperformed the lemmatized version (F1-score: 0.9317; accuracy: 0.9320), likely due to the informal nature of user-generated content. Among the SVM kernels tested, the Radial Basis Function (RBF) kernel achieved the best performance overall (F1-score: 0.9287; accuracy: 0.9290). Furthermore, the use of Chi-Square feature selection consistently improved model performance by reducing feature noise and increasing classification accuracy. This finding highlights the importance of carefully selecting preprocessing methods and feature selection strategies tailored to the dataset domain. Overall, the combination of non-lemmatized preprocessing, TF-IDF, Chi-Square feature selection, and the RBF kernel offers a robust pipeline for effective sentiment classification of game reviews on digital platforms. This study is one of the first to systematically examine the combined impact of preprocessing and feature selection techniques on SVM- based sentiment classification of Steam game reviews. The approach can also be applied in automatic game quality assessment systems to help developers monitor player satisfaction trends.