LOREN is an interpretable fact verification model trained on FEVER, which aims to predict the veracity of a textual claim against a trustworthy knowledge source such as Wikipedia. LOREN also decomposes the verification and makes accurate and faithful phrase-level veracity predictions without any phrasal veracity supervision.
This repo hosts the pre-trained fact verification model for LOREN based on RoBERTa (large). More technical details can be found at this GitHub Repo.
Please check out our AAAI 2022 paper for more details: "LOREN: Logic-Regularized Reasoning for Interpretable Fact Verification".