<![CDATA[This study aimed to develop a hoax news detection application in the Indonesian language using the IndoBERT model, implemented through the Streamlit platform. The increasing spread of false information in Indonesia serves as the main motivation behind the development of this system. The classification model was trained using a partial fine-tuning strategy, where most of the early encoder layers were frozen while the top layers and classifier head were retrained. The dataset consisted of 24,658 preprocessed and summarized news entries using the T5-base model, with the main features being the combination of headlines and summarized content. The best-performing model achieved an accuracy of 97.81%, precision of 98.31%, recall of 97.56%, F1-score of 97.74%, and an ROC AUC score of 0.996. The selected model was then integrated into a Streamlit-based application, enabling users to input news text and receive real-time classification results of “Hoax” or “Not Hoax” along with the confidence score. The application was designed with a simple and user-friendly interface and aimed to assist general users in verifying information independently. The results demonstrate that the combination of IndoBERT and Streamlit is effective for building an accurate and practical hoax detection system.]]>
Copyrights © 2026
