Development and Evaluation of an IndoBERT-Based NLP Model for Automated Clickbait Detection

Abstract

The rapid growth of digital news platforms necessitates reliable and automated systems for maintaining content quality at scale. This study presents the engineering and evaluation of an IndoBERT-based Natural Language Processing (NLP) framework for automated clickbait detection in Indonesian news headlines. The proposed framework is designed as an end-to-end text classification pipeline, incorporating data preprocessing, tokenization, fine-tuning of a pretrained IndoBERT model, and systematic performance evaluation. Experiments were conducted using the CLICK-ID dataset comprising 15,000 Indonesian news headlines, with an 80:20 stratified train–test split. The fine-tuned model achieved an accuracy of 0.83, with a precision of 0.82, recall of 0.77, and an F1-score of 0.79 for the clickbait class. Further evaluation using threshold-independent metrics yielded a ROC-AUC value of 0.89 and an average precision of 0.88, indicating strong discriminative capability under moderate class imbalance. Comparative analysis shows that the proposed approach outperforms prior CNN, Bi-LSTM, and ensemble-based methods evaluated on the same dataset. These results demonstrate that IndoBERT provides a robust foundation for engineering automated clickbait detection systems tailored to Indonesian-language news streams.