Analisis Sentimen Pada Media Sosial X Terhadap Akun Fiktif Menggunakan IndoBERT

Social media has become a platform where people interact and express opinions openly. One phenomenon that gained attention on social media X is the appearance of a fictitious account named Fufufafa, which sparked controversy through its old provocative posts targeting public figures on Kaskus. To understand public responses to this issue, sentiment analysis is needed to capture the variety of expressions in the Indonesian language. Previous studies mostly used traditional methods such as SVM and rarely applied transformer-based approaches like IndoBERT, especially with data augmentation to handle imbalanced data. This study aims to identify public sentiment toward the Fufufafa account on social media X and evaluate the effect of synonym replacement and back translation techniques on IndoBERT performance. The research includes data collection through crawling, preprocessing, data splitting, and model training in three scenarios: baseline, synonym replacement, and back translation. The results show that the back translation scenario with a learning rate of 0.00003 achieved the best performance with an average precision, recall, and F1-score of 65% and accuracy of 84%. In addition, a contextual analysis of the negative class was conducted to identify utterance types such as hate speech, offensive, and toxic to understand public opinion more deeply.