rishhh/verisci-claim-verifier-retrieval-aware-seed123

VeriSci Claim Verifier Retrieval-Aware Candidate

Retrieval-aware cross-encoder/nli-deberta-v3-base classifier for scientific claim verification over SUPPORTS, REFUTES, and NOT_ENOUGH_INFO.

Training Data

• —allenai/scifact_entailment train split with gold evidence sentences.
• —allenai/scifact train claims converted into gold evidence snippets, BM25-selected evidence snippets, and BM25-retrieved hard NOT_ENOUGH_INFO negatives.
• —andreiaalexa/scifact-relevance-pairs title/train hard negatives.

Derived Test

Accuracy: 0.8757 Macro F1: 0.8353

Full SciFact Validation Gate

Top-1 end-to-end accuracy: 0.5356 Top-1 end-to-end macro F1: 0.5154 Top-5 decisive accuracy: 0.6044 Top-5 decisive macro F1: 0.5712

See evaluation/eval_summary.json for full metrics and limitations.

Responsible Use

This model is not a medical device, not a substitute for peer review, and should not be used for clinical, legal, or public-policy decisions without expert review.