FactReview: Evidence-Grounded Peer Review with Execution-Based Claim Verification
Researchers have developed FactReview, a system designed to enhance the peer-review process for machine learning papers by verifying empirical claims. FactReview extracts claims from manuscripts, grounds them in related work, and crucially, executes code artifacts to audit these claims. In evaluations across 35 papers, FactReview covered 84% of claims and significantly improved review quality and efficiency, reducing reviewer time by 58% while increasing claim coverage. AI
IMPACT This system could significantly improve the rigor and efficiency of scientific peer review, particularly in fields like machine learning where code execution is critical for verifying claims.