Deteksi Berita Hoaks dari TurnBackHoax.id Menggunakan Algoritma BERT

Abstract

The widespread circulation of misleading information on digital platforms has become a critical issue, as it can distort public understanding and influence decision-making processes. TurnBackHoax.id serves as a reference platform for information verification by providing curated collections of both false and verified news to support digital literacy initiatives. This study aims to detect hoax-related news by applying the Bidirectional Encoder Representations from Transformers (BERT) algorithm, utilizing a dataset obtained from TurnBackHoax.id. The research methodology encompasses data collection, text preprocessing, data annotation, BERT model training, and performance evaluation. The experimental results demonstrate that the BERT model is capable of effectively distinguishing between hoax and non-hoax news, achieving an accuracy rate of 86%, along with high precision, indicating strong performance in capturing the contextual characteristics of the Indonesian language. Furthermore, an analysis loss suggests that the model exhibits good generalization ability without significant signs of overfitting. These findings are expected to provide a foundation for the development of automated hoax detection systems based on deep learning techniques to support digital literacy efforts and combat misinformation in Indonesia.