This is an implemetation of the paper Few-shot Relation Extraction via Bayesian Meta-learning on Relation Graphs.
The codes rely on pre-trained BERT models. Please download pretrain.tar from Tsinghua Cloud and put it under the root. Then run tar xvf pretrain.tar to decompress it.
To run the model on the FewRel dataset, we could use the following command:
python train_demo.py --trainN 5 --N 5 --K 1 --Q 1 --model regrab --encoder bert --hidden_size 768 --val_step 1000 --batch_size 8 --fp16 --seed 1
Most of the codes are from the FewRel repo, which provides a neat codebase for few-shot relation extraction.
Please consider citing the following paper if you find our codes helpful. Thank you!
@inproceedings{qu2020few,
title={Few-shot Relation Extraction via Bayesian Meta-learning on Relation Graphs},
author={Qu, Meng and Gao, Tianyu and Xhonneux, Louis-Pascal AC and Tang, Jian},
booktitle={International Conference on Machine Learning},
year={2020}
}