Developing an Indonesian Fake News Detection System Using IndoBERT and ProtoNet for Few-Shot Learning

Fake news dissemination through digital media has become a critical concern because inaccurate information may spread swiftly and influence public comprehension, social behavior, and decision-making. This study developed an Indonesian fake news detection web application employing fine-tuned IndoBERT and an IndoBERT-based Prototypical Network (ProtoNet) under few-shot learning scenarios. The research followed the Cross Industry Standard Process for Data Mining framework, consisting of business understanding, data understanding, data preparation, modeling, evaluation, and deployment. The dataset was obtained from the publicly available Deteksi Berita Hoaks Indo Dataset on Kaggle and contained 23,944 Indonesian news records, consisting of 11,200 factual and 12,744 hoax articles. Three few-shot scenarios were evaluated: 50-shot, 100-shot, and 300-shot per class. IndoBERT was fine-tuned as a supervised classification baseline, while ProtoNet used IndoBERT embeddings to construct class prototypes and classify texts based on Euclidean distance. The best performance was achieved by fine-tuned IndoBERT in the 300-shot scenario, with 97.12% accuracy and a 97.10% macro F1-score. ProtoNet achieved 92.65% accuracy and a 92.63% macro F1-score in the same scenario. Even in the most constrained 50-shot setting, ProtoNet achieved an F1-score of 90.82%, closely trailing fine-tuned IndoBERT at 91.17% and demonstrating strong metric stability. The web application supports manual text input, URL-based article extraction, model selection, confidence scores, and word-level occlusion explanations. These results demonstrate the effectiveness of IndoBERT for Indonesian fake news detection and the feasibility of prototype-based classification under limited-data conditions. Overall, this work bridges the gap between deep contextual representation, data-efficient learning, and practical, interpretable deployment for public media verification.