Skip to content

Latest commit

 

History

History
68 lines (47 loc) · 2.02 KB

README.md

File metadata and controls

68 lines (47 loc) · 2.02 KB
Raw

MVRE

Code for the AAAI 2024 long paper "Enhancing Low-Resource Relation Representations through Multi-View Decoupling"

Model Architecture

Requirements

To install requirements:

pip install -r requirements.txt

How to run

Initialize the answer words

Use the comand below to get the answer words to use in the training.

python get_label_word.py --model_name_or_path roberta-large  --dataset_name semeval

The {answer_words}.ptwill be saved in the dataset, you need to assign the model_name_or_path and dataset_name in the get_label_word.py.

Split few-shot dataset

Download the data first, and put it to dataset folder. Run the comand below, and get the few shot dataset.

python generate_k_shot.py --data_dir ./dataset --k 5 --dataset semeval
cd dataset
cd semeval
cp rel2id.json val.txt test.txt ./k-shot/5-1

You need to modify the k and dataset to assign k-shot and dataset. Here we default seed as 1,2,3,4,5 to split each k-shot, you can revise it in the generate_k_shot.py

Run

bash scripts/semeval.sh 
bash scripts/tacred.sh
bash scripts/tacrev.sh

Acknowledgement

Part of our code is borrowed from code of RetrievalRE, many thanks.

Papers for the Project & How to Cite

If you use or extend our work, please cite the paper as follows:

@article{Fan_Wei_Qu_Lu_Xie_Cheng_Chen_2024,
      title={Enhancing Low-Resource Relation Representations through Multi-View Decoupling}, 
      volume={38}, url={https://ojs.aaai.org/index.php/AAAI/article/view/29752}, 
      DOI={10.1609/aaai.v38i16.29752}, 
      number={16}, 
      journal={Proceedings of the AAAI Conference on Artificial Intelligence}, 
      author={Fan, Chenghao and Wei, Wei and Qu, Xiaoye and Lu, Zhenyi and Xie, Wenfeng and Cheng, Yu and Chen, Dangyang}, year={2024}, month={Mar.}, pages={17968-17976}
}