PRGC: Potential Relation and Global Correspondence Based Joint Relational Triple Extraction

This repository contains the source code and dataset for the paper: PRGC: Potential Relation and Global Correspondence Based Joint Relational Triple Extraction. Hengyi Zheng, Rui Wen, Xi Chen et al. ACL 2021.

Overview

Requirements

The main requirements are:

• python==3.7.9
• pytorch==1.6.0
• transformers==3.2.0
• tqdm

Datasets

• NYT* and WebNLG*(following CasRel)
• NYT(following CopyRE)
• WebNLG(following ETL-span)

Or you can just download our preprocessed datasets.

Usage

1. Get pre-trained BERT model for PyTorch

Download BERT-Base-Cased which contains pytroch_model.bin, vocab.txt and config.json. Put these under ./pretrain_models.

2. Build Data

Put our preprocessed datasets under ./data.

3. Train

Just run the script in ./script by sh train.sh.

For example, to train the model for NYT* dataset, update the train.sh as:

python ../train.py \
--ex_index=1 \
--epoch_num=100 \
--device_id=0 \
--corpus_type=NYT-star \
--ensure_corres \
--ensure_rel

4. Evaluate

Just run the script in ./script by sh evaluate.sh.

For example, to train the model for NYT* dataset, update the evaluate.sh as:

python ../evaluate.py \
--ex_index=1 \
--device_id=0 \
--mode=test \
--corpus_type=NYT-star \
--ensure_corres \
--ensure_rel \
--corres_threshold=0.5 \
--rel_threshold=0.1