Analisis Sentimen dan Identifikasi Topik Kritik Pembaruan Game Umamusume Menggunakan SVM Dan LDA

Abstract

Whenever a new update or patch is released for a gacha game, the player community typically responds in various ways. Consequently, this study focuses on sentiment analysis and the identification of key topics on the social media platform X regarding updates to the game *Umamusume: Pretty Derby*. Data was collected via scraping, resulting in 139 cleaned data points after a preprocessing phase. Following TF-IDF weighting, the data was classified using a Support Vector Machine (SVM) model, which achieved an accuracy of 64.29%. Evaluation results indicate that the model performs reasonably well in identifying positive and neutral sentiments; however, it struggles to detect negative sentiments, partly due to an imbalanced dataset and the frequent use of sarcasm. Meanwhile, the Latent Dirichlet Allocation (LDA) method successfully identified three primary topics: reactions to patch notes, and discussions concerning characters and gameplay. It is hoped that these findings will provide valuable input for developers to evaluate future updates. Although further improvements are needed, this level of accuracy demonstrates the potential of the SVM approach for text classification within the gacha game domain.