Computational Fact-checking

Journalists cannot read every transcript in full. We build resources that help, from a benchmark dataset of check-worthy factual claims to ClaimViz, which we believe is the first visual analytics system for verifying claims in spoken text such as debates, speeches and interviews.

ClaimViz lets fact-checkers filter a long transcript down to its most check-worthy sentences, then find evidence that supports or contradicts each claim. Recent work surveys information disorder on video-sharing platforms and detects misinformation that combines text, images and video.