Local explainability-based model for clickbait spoiler generation

Clickbait involves creating attention-grabbing or deceptive content aimed at generate more clicks. While effective for driving online traffic, it often results in misinformation, user frustration, and a diminished experience. Therefore, promptly identifying and addressing clickbait is vital. Spoiling clickbait involves crafting concise messages that divulge the actual content, serving as a means to counter its effects. Creating brief messages that expose the genuine nature of clickbait posts is one method of countering it. The proposed model first identifies the clickbait highlighting the words using the Local Interpretable Model-Agnostic Explanations (LIME) approach which helps to mark it as clickbait. The sentences containing those highlighted words are extracted from a dataset to create a spoiler for that post. The Bilingual Evaluation Understudy (BLEU) score of 0.61 and Recall-Oriented Understudy for Gisting Evaluation (ROUGE) score of 0.72 obtained from the proposed spoiler generation model surpassed the prior state-of-the-art models.