Benchmarking Clickbait Detection from News Headlines

To capture reader attention, news media increasingly resort to sensationalist headlines-a practice commonly referred to as clickbait. To assess the prevalence of clickbait in Taiwan news media and develop an effective detection model, we labeled a dataset and tested various models, including Support Vector Machine (SVM), Random Forest (RF), and Bidirectional Encoder Representations from Transformers (BERT). The CKIP-BERT model, pre-trained by Yang and Ma, outperformed the others, achieving F-scores of 0.887 in binary classification and 0.918 in multi-class classification, establishing a new benchmark for clickbait detection in Taiwan news headlines.