Fake News Detection in Indonesian Language Using a Deep Learning Approach with Indo-BERT

Abstract

The development of digital technology and social media has increased people's exposure to information, but it also raises serious challenges in the form of the spread of fake news. This research aims to develop an Indonesian fake news detection system by utilizing the Indo-BERT model with a Deep Learning approach, a transformer model that has been trained using a large corpus of Indonesian language. The research dataset consists of thousands of articles from CNN Indonesia, Kompas, and Tempo as the original news, and TurnBackHoax as the source of fake news. After going through the text pre-processing stage, the Indo-BERT model was fine-tuned for binary classification. The test results showed excellent performance with an accuracy of 92%, F1-score of 0.92, and ROC value of 0.98, confirming the model's ability to consistently distinguish between real news and hoaxes. The trained model is then integrated into a Web-based application, so that it can be used directly by the public to verify news. In addition to making technical contributions in the Natural Language Processing (NLP) domain, this research also emphasizes social and educational dimensions, namely supporting digital literacy, increasing critical awareness, and strengthening technology-based learning strategies in dealing with misinformation.