Reproduce “LLM-based Corroborating and Refuting Evidence Retrieval for 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
Siyuan Wang, James R. Foulds, Md Osman Gani, Shimei Pan
CIBER evaluates scientific claims by adding agreeing, conflicting, and paraphrased questions about retrieved papers, then combining model responses into support, refute, or neutral verdicts and confidence scores. Across four balanced abstract datasets covering two claims, GPT-3.5-Turbo with CIBER achieved 0.6667 accuracy and 0.6225 F1, versus 0.5188 and 0.4857 with conventional RAG. These scores measure classification, not hallucination reduction or broad generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.