ClickGuard browser extension detects clickbait with 91% F1-score, auto-generates spoilers
A new browser extension uses transformer embeddings and XGBoost to catch clickbait with 91% accuracy...
ClickGuard is a new AI-powered browser extension designed to combat clickbait by detecting misleading headlines and providing instant spoilers. Developed by researchers Wojciech Michaluk, Tymoteusz Urban, Mateusz Kubita, Soveatin Kuntur, and Anna Wróblewska, the system uses a hybrid machine learning architecture that combines transformer-based embeddings (e.g., from LLMs) with linguistically motivated features and a custom 'baitness' score. After evaluating multiple NLP techniques—from classic vectorizers to modern LLM embeddings—the team settled on an XGBoost-based model that achieves a 91% F1-score on an open combined dataset.
The extension operates in two phases: it warns users before they click a suspicious link, and after opening an article it displays a percentage score indicating the likelihood of clickbait. Each prediction is explained using analyzed metrics, including the novel measures proposed in the system. Perhaps its most striking feature is the clickbait spoiler: a one- to two-sentence summary of the entire article, letting users get the gist without falling for sensationalism. A demo video accompanies the paper. This practical tool directly addresses information hygiene on the web, moving beyond simple classification to deliver transparent, actionable insights for everyday browsing.
- Hybrid ML pipeline: transformer embeddings + linguistic features + custom 'baitness' score.
- XGBoost model achieves 91% F1-score on combined open clickbait dataset.
- Generates one- to two-sentence spoiler summaries plus per-article percentage likelihood scores with explanations.
Why It Matters
ClickGuard puts AI-powered clickbait detection and spoilers directly in your browser, reducing misinformation exposure.