aiml-clickbait-benchmark

Clickbait detection evaluates headlines using a BERT-based classifier against a fixed set of misinformation domains. It validates domain coverage, body length, and headline sensationalism, reporting counts and failure rates for missing domains or placeholders.

To benchmark a BERT classifier, run the skill on a local headlines.json file containing articles with misinfo_domain, headline, and body fields. It labels headlines and reports detection rates, failing if required misinformation domains are missing.

Yes, you can validate dataset coverage by checking articles against a fixed MISINFO_DOMAINS set. The benchmark fails if any required misinformation domains are absent or if dataset placeholders are detected.

Automated clickbait benchmarking requires a local headlines.json file and a dataset containing specific fields like misinfo_domain, headline, and body. It will fail if domains are missing, placeholders exist, or body length and sensationalism thresholds are not met.

Your clickbait benchmark fails when the supplied articles dataset does not cover all domains in the fixed MISINFO_DOMAINS set, or when placeholder text is detected, ensuring dataset realism before moderation model deployment.