Emotion-Aware Clickbait Attack in Social Media

A new research paper from Syed Mhamudul Hasan, Mohd. Farhan Israk Soumik, and Abdur R. Shahid proposes an emotion-aware clickbait generation attack that systematically optimizes emotional impact to evade existing detection systems. The framework models clickbait dynamics using the Valence-Arousal-Dominance (VAD) space, a psychological model that captures emotional states along three dimensions: pleasure, arousal, and dominance. By leveraging Large Language Models (LLMs) to generate multiple stylistic rewrites of headlines, and Sentence-BERT to semantically align them with real social media posts, the attack creates clickbait that appears natural but carries disproportionately high emotional intensity.

The paper introduces a Curiosity Gap (CG) function that measures how a headline's emotional activation deviates from an authentic post, quantifying the user engagement potential while evading surface-level detectors. Experimental results show that emotion-aware stylization degrades the performance of state-of-the-art classifiers, causing misclassification rates between 2.58% and 30.63%—a significant blind spot in current clickbait detection. This work highlights that clickbait is not merely a static textual phenomenon but a dynamic emotional manipulation, and calls for detection systems to incorporate emotional dynamics rather than relying solely on linguistic patterns or structural cues.