Analisis Sentimen Pemain terhadap Sistem Gacha Genshin Impact di Media Sosial X Menggunakan SVM

Abstract

The gacha system in Genshin Impact allows players to randomly obtain rare characters or weapons, triggering diverse emotional responses on social media platform X. This study aims to analyze player sentiment trends toward the gacha system and evaluate the performance of the Support Vector Machine (SVM) algorithm in classifying such sentiment. The research data consisted of 690 Indonesian-language tweets collected through web scraping using the Tweet Harvest library, which were then processed through several preprocessing stages (cleaning, case folding, tokenizing, stopword removal, and stemming) and labeled into positive, negative, and neutral sentiment classes. Text features were extracted using TF-IDF, producing a 690×3440 weighted matrix, and classified using a linear-kernel SVM with a class_weight='balanced' parameter and an 80:20 data split. Labeling results showed a dominance of neutral sentiment (65.94%), followed by positive (18.41%) and negative (15.65%). The SVM model achieved 74% accuracy on the test data; however, further evaluation revealed significant overfitting, indicated by a 44.37% gap in Macro Average F1-Score between training data (99.35%) and testing data (54.97%). The neutral class exhibited the best performance (97% Recall), while the negative and positive classes showed low Recall (23% and 36%, respectively) due to vocabulary overlap between informal expressions and the neutral class. These findings suggest that SVM performance on informal Indonesian-language microblogging data is more susceptible to overfitting compared to more structured review data.