<![CDATA[The rapid spread of misinformation in digital media has become a serious challenge, particularly in the context of Indonesian online news consumption. Therefore, this study aims to develop an effective fake news detection system by leveraging a contextual language model tailored to the Indonesian language. This study proposes the use of the Indonesian Bidirectional Encoder Representations from Transformers (IndoBERT) model to detect fake news by optimizing the model architecture and hyperparameter settings. IndoBERT, as a contextual language model, functions to capture the nuances of language in text and can be adjusted at the output layer for classification tasks. In this study, the model architecture is enhanced by adding classification layers, including a linear layer, ReLU activation function, dropout, and a sigmoid output layer, which is expected to improve the accuracy of fake news detection. This study uses Indonesian news data from two sites, Turnbackhoax and Detik.com, covering various topics. The model is trained with different hyperparameter settings, including batch size, dropout, and learning rate, to find the best combination that achieves high accuracy in detecting fake news. The trained model is integrated into a Flask-based web application, allowing users to automatically verify the truth of news through a simple and user-friendly interface. The results show that the best combination, with a learning rate of 2e-5, dropout of 0.1, and a batch size of 32, achieves an accuracy of 96.4%. This study demonstrates that IndoBERT, with the appropriate settings, can be effectively applied to detect fake news with high accuracy, contributing to the development of hoax detection applications.]]>
Copyrights © 2026
