Reproduce “ScienceJudge: Pattern Perturbation Augmentation for Sentence-Level Scientific Claim Verification” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Chuhai Cai, Xin Chen, Min Yang, Feng Jiang

ScienceJudge targets scientific claim errors by learning label-specific edits from observed reporting errors and applying them to supported statements, creating hard negatives. It combines PPA-trained vision-language models with instruction-guided language-model predictors. On a reused development set, its average score is 0.8692, versus 0.8124 for Base Training. Its first-place official result was 0.6890; the local result is not independent confirmation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.