Sistem Deteksi Berita Hoaks berbasis Algoritma Natural Language Processing (NLP) menggunakan BERT

The advancement of digital technology brings both benefits and challenges, one of which is the increasing spread of hoaxes that can trigger conflicts in various sectors such as social, cultural, political, and economic. Hoaxes are unverifiable and often provocative information that spreads rapidly across digital platforms. Indonesian society remains vulnerable to unverified information. Therefore, an artificial intelligence (AI)-based system is needed to automatically detect hoaxes. This study employs the BERT model for its ability to understand word context and perform effective semantic classification through tokenization and transformer architecture. The dataset, sourced from Kaggle, consists of 730 articles: 425 labeled as hoax and 305 as non-hoax. After preprocessing and tokenization, the data was input into the model. BERT was chosen for its strong word representation capabilities trained on a large-scale corpus. Evaluation results show that BERT achieved 98% accuracy, outperforming the previous KNN model which reached 93.33%. These findings demonstrate the effectiveness of the BERT-based approach in detecting digital disinformation.