CLICK-ID MULTI: A Multimodal Dataset for Indonesian Clickbait Detection and Benchmarking

ACL ARR 2025 May Submission1449 Authors

Abstract: Clickbait headlines attract user attention by exploiting curiosity gaps, often through sensational or misleading phrasing, while not necessarily conveying false information. Although clickbait contributes to the broader misinformation ecosystem, especially when amplified on social media, it remains underexplored in low-resource and multimodal settings. This paper introduces CLICK-ID MULTI, a new multimodal dataset for clickbait detection in Indonesian. It extends the original CLICK-ID dataset (William and Siri, 2020) by pairing 5,809 annotated news articles with associated images, enabling the development of multimodal models. Despite its smaller size compared to the original text-only dataset, CLICK-ID MULTI supports models that outperform the best text-only baseline (F1 = 0.7365), achieving F1 scores up to 0.937 through image-text fusion. These findings highlight the importance of multimodal learning and language-specific pretraining for robust clickbait detection in low-resource languages. The dataset and code are publicly available at: https://anonymous.4open.science/r/emnlp-2025-clickid-multi-8466.

Paper Type: Short

Research Area: Computational Social Science and Cultural Analytics

Research Area Keywords: misinformation detection and analysis

Contribution Types: NLP engineering experiment

Languages Studied: Indonesian

Submission Number: 1449