Source code for "Document-Level Relation Extraction with Relation Correlation Enhancement", International Conference on Neural Information Processing, 2023.
- Python 3.9.13
- CUDA 12.0
- transformers 4.11.3
- torch 1.13.1
- DocRED dataset can be downloaded following their instructions
- These files should be placed in the following format
Dataset
|- DocRED
|- train_annotated.json
|- train_distant.json
|- dev.json
|- test.json
Train LACE model using the following command,
>> CUDA_VISIBLE_DEVICES=0 nohup sh run_bert.sh >train.log 2>&1 &
Experimental results are verified by transmitting the output of the test set to CodaLab.
Registration is required for testing.
If you find our work inspiring, please kindly cite the following paper,
@article{huang2023document,
title={Document-Level Relation Extraction with Relation Correlation Enhancement},
author={Huang, Yusheng and Lin, Zhouhan},
journal={arXiv preprint arXiv:2310.13000},
year={2023}
}