Unsupervised Fact Error Correction Modeling by Using Span-Level Contrastive Learning

Source code for our paper : Unsupervised Fact Error Correction Modeling by Using Span-Level Contrastive Learning.

Click the links below to view our checkpoints

Requirement

1. Install the following packages using Pip or Conda under this environment

Python==3.9
Pytorch
transformers

We provide the version file requirements.txt of all our used packages, if you have any problems configuring the environment, please refer to this document.

Reproduce CorrectFEC

Download Code & Dataset

• First, use git clone to download this project:

git clone https://github.com/NEUIR/CorrectFEC
cd CorrectFEC

• Download link for data and model

Train CorrectFEC

I will show you how to reproduce the results in the CorrectFEC paper.

• Go to the model folder and train the CorrectFEC model checkpoint:

cd model
bash train.sh

Inference CorrectFEC

• For the FEVER and SCIFACT dataset: Go to the inference folder and inference on the CorrectFEC model:

cd inference
bash inference_final.sh

Evaluate Prediction Effectiveness

• These experimental results are shown in Table 1 of our paper.
• Go to the evals folder and evaluate model performance as follow:

cd evals
bash evals.sh
python chatgpt_gpteval.py
python sentencebert.py

Contact

If you have questions, suggestions, and bug reports, please email:

[email protected]