<![CDATA[Abstract The spread of misinformation (hoaxes) on social media, particularly on the X platform, has become a significant challenge that can influence public understanding and perceptions of various issues, including those related to education. This study aims to analyze and improve the performance of an Indonesian-language hoax detection model through a fine-tuning approach using the IndoBERT model. The dataset used in this research was collected through web scraping from the X platform and classified into two categories: hoax and non-hoax. Prior to model training, the data underwent several preprocessing stages, including data cleansing, normalization, tokenization, stopword removal, and stemming. This study utilized the pre-trained IndoBERTweet model with two data-splitting scenarios, namely 70:30 and 80:20, along with various combinations of learning rates and epochs to determine the optimal configuration. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. The experimental results indicate that the best performance was achieved under the 70:30 data-splitting scenario with a learning rate of 2e-5 and 20 epochs. On the testing dataset, the model achieved an accuracy of 80%, precision of 83%, recall of 78%, and an F1-score of 80%. The findings demonstrate that the fine-tuning approach applied to IndoBERT is effective in enhancing the performance of Indonesian-language hoax classification and is capable of maintaining strong performance when evaluated on previously unseen data. Keywords: IndoBERT, IndoBERTweet, fine-tuning, hoax detection, social media, natural language processing (NLP), text classification.]]>
Copyrights © 2026
