GitHub - thunlp/RE-Context-or-Names: Bert-based models(BERT, MTB, CP) for relation extraction. (original) (raw)

RE Analysis

Dataset and code for Learning from Context or Names? An Empirical Study on Neural Relation Extraction.

If you use this code, please cite us

@article{peng2020learning,
  title={Learning from Context or Names? An Empirical Study on Neural Relation Extraction},
  author={Peng, Hao and Gao, Tianyu and Han, Xu and Lin, Yankai and Li, Peng and Liu, Zhiyuan and Sun, Maosong and Zhou, Jie},
  journal={arXiv preprint arXiv:2010.01923},
  year={2020}
}

Quick Start

You can quickly run our code by following steps:

• Install dependencies as described in following section.
• cd to pretrain or finetune directory then download and pre-processing data for pre-traing or finetuning.

1. Dependencies

Run the following script to install dependencies.

pip install -r requirement.txt

You need install transformers and apex manually.

transformersWe use huggingface transformers to implement Bert. And for convenience, we have downloaded transformers into utils/. And we have also modified some lines in the class BertForMaskedLM in src/transformers/modeling_bert.py while keep the other codes unchanged.

You just need run

to install transformers manually.

apexInstall apex under the offical guidance.

2. More details

You can cd to pretrain or finetune to learn more details about pre-training or finetuning.