shilida / CL_Interactive_Argument_Pair_Identification

A Simple Contrastive Learning Framework for Interactive Argument Pair Identification via Argument-Context Extraction(EMNLP 2022)

The method is implemented using PyTorch.

Folder

• config: the config and hyperparameter file
• data: please put data in this folder
• dataset.py: data preprocessing file
• models.py: model file.
• evaluation.py: evaluation function file
• train.py: training, validation function file
• test.py: testing function
• ace.py: argument-context extraction module

Train a model:

python train.py

config/hyparameter.json is the config file. It contains a number of hyperparameters. Hyperparameters can be modified for custom training.
noisy: NO,RandomWordAug,BackTranslationAug,KeyboardAug
objective:BCE,BCE+SCL
hard_sample_con:NO,YES
model_type：bert_without_context,bert_with_context

Test a model:

python test.py

Contact

[e-mail]

Citation

@inproceedings{shi2022simple,
title={A Simple Contrastive Learning Framework for Interactive Argument Pair Identification via Argument-Context Extraction},
author={Shi, Lida and Giunchiglia, Fausto and Song, Rui and Shi, Daqian and Liu, Tongtong and Diao, Xiaolei and Xu, Hao},
booktitle={Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing},
pages={10027--10039},
year={2022}
}