More than Classification: A Unified Framework for Event Temporal Relation Extraction

This is the repository for the resources in ACL 2023 Paper "More than Classification: A Unified Framework for Event Temporal Relation Extraction". This repository contains the source code and datasets used in our paper.

Dataset

Two datasets (MATRES and TBDense) are used for training in the paper.

Since the previous URL link has expired, we uploaded the datasets to the ./data file.

How to train

Environment Setup et al.

git clone https://github.com/AndrewZhe/A-Unified-Framework-for-Event-Temporal-Relation-Extraction.git
conda create -n conda-env python=3.7.9
conda activate conda-env
pip install -r requirements.txt
mkdir MATRES/json
mkdir TBDense/json
mkdir log
mkdir models

Prepare Data

Process the raw MATRES/TBDense data.

The processed data will be under data/<TASK>/json/

python run_data.py --task <TASK>

<TASK>: choose from "TBDense", "MATRES"

Training and Testing

To train the model with MATRES

bash ./scripts/run_matres_bert.sh

bash ./scripts/run_matres_roberta.sh

To train the model with TBDense

bash ./scripts/run_tbdense_bert.sh

bash ./scripts/run_tbdense_roberta.sh

--use_time_point: 0 for baseline, 4 for our model

--task: choose from "`TBDense", "MATRES"

--temperature: used to control the smoothing degree of the probability distribution

--learning_rate: learning rate for Classifier

--bert_lr: learning rate for Encoder

--num_train_epochs: number of training epochs. 1 for BERT-Base, 3 for RoBERTa-Large.

--bert_model_dir: dir of PLM BERT-Base or RoBERTa-Large

The model will be saved in ./models/
The training and testing result will be reported in log: ./log/

Reference

Bibtex:

@misc{huang2023classification,
      title={More than Classification: A Unified Framework for Event Temporal Relation Extraction}, 
      author={Quzhe Huang and Yutong Hu and Shengqi Zhu and Yansong Feng and Chang Liu and Dongyan Zhao},
      year={2023},
      eprint={2305.17607},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

If your use MATRES, please refer:

@inproceedings{ning2018multi,
  title={A Multi-Axis Annotation Scheme for Event Temporal Relations},
  author={Ning, Qiang and Wu, Hao and Roth, Dan},
  booktitle={Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  pages={1318--1328},
  year={2018}
}

If your use TBDense, please refer:

@inproceedings{cassidy2014annotation,
  title={An Annotation Framework for Dense Event Ordering},
  author={Cassidy, Taylor and McDowell, Bill and Chambers, Nathanael and Bethard, Steven},
  booktitle={Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)},
  pages={501--506},
  year={2014}
}