This repository hosts data and code used in the paper 'Aggregating Pairwise Semantic Differences for Few-Shot Claim Verification'.
Installing a dedicated Python environment is recommended with the following code:
conda env create -n seed --file environment.yml
fever contains the data used for three-way claim verification on FEVER. It is the original test set of the FEVER datset.
fever2 contains the data used for binary claim verification on FEVER. It is kindly offered by the authors of 'Towards Few-Shot Fact-Checking via Perplexity'.
scifact contains the data used for three-way calim verification on SCIFACT. It is the original files of the SCIFACT dataset.
We apply SEED to binary FEVER, three-way FEVER, three-way SCIFACT with seed_fever_binary, seed_fever and seed_scifact respectively.
To reproduce the experiments reported in the paper, use run_seed.sh with specified task name at the beginning. 'bfever' stands for binary FEVER; 'fever' stands for three-way FEVER; 'scifact' stands for three-way SCIFACT.
To reproduce the finetuning experiments reported in the paper, use run_finetuning.sh with specified task name at the beginning. 'bfever' stands for binary FEVER; 'fever' stands for three-way FEVER; 'scifact' stands for three-way SCIFACT.