Analisis Sentimen terhadap Kebijakan Naturalisasi Timnas Indonesia Menggunakan IndoBERT dengan Perbandingan Augmentasi WordNet dan SMOTE

The naturalization of players for the Indonesian National Football Team has sparked diverse public opinions on social media platform X (Twitter), ranging from support for improving team quality to concerns about the marginalization of local players. To analyze public sentiment more comprehensively, this study employs the IndoBERT model with two data balancing approaches: WordNet-based augmentation (synonym replacement and random deletion) and the Synthetic Minority Oversampling Technique (SMOTE). A total of 7,866 tweets were collected through a crawling process using the keyword “naturalisasi Timnas Indonesia” (“Indonesian national team naturalization”). The results indicate that both balancing methods successfully improved the model’s performance, particularly for the Neutral class, which previously had the lowest F1-score. The WordNet + IndoBERT model achieved an accuracy of 83% with a macro average F1-score of 0.77, while SMOTE + IndoBERT obtained an accuracy of 81% and a macro average F1-score of 0.75. These findings suggest that both methods are effective in reducing bias toward the majority class, with WordNet performing slightly better due to its ability to enrich the semantic meaning of text. However, the model still faces challenges in distinguishing Neutral and Positive sentiments, likely due to the linguistic similarities and diverse expressions of public opinion on social media.