ERECT: Evidence Refinement Enhanced Complex Claim Verification with Large Language Models

The contributions of this paper are as follows:
– Introduction of ERECT model: We propose a novel evidence refinement enhanced complex claim verification model, ERECT, which effectively decomposes complex claim verification into simpler program steps and get refined evidence to support the excution of simpler program steps.
– Integration of LLM for evidence refinement: Our approach leverages large language models (LLMs) to refine evidence from a large external corpus, ensuring that the most relevant evidence is selected in evidence retrieval step. This refinement significantly enhances the precision of the claim verification process.
– Evaluation of the importance of evidence: We designed ablation experiments to test the performance of the same model under different evidence type settings, quantifying the importance of evidence accuracy in the FV task.