Code for the paper "BERT-enhanced Relational Sentence Ordering Network" (EMNLP 2020).
We choose the ROC dataset as an example.
Besides, we clean the original code and remove some useless modules which contribute little to the overall performance.
python 3.6.5, pytorch 1.1.0
The recommended hyperparameters for the current code of BERSON.
ROC dataset: batch=4, lr=2e-5, epoch=2, coefficient=0.6
AAN dataset: batch=4, lr=5e-5, epoch=2, coefficient=0.4
NIPS dataset: batch=8, lr=5e-5, epoch=20, coefficient=0.1
SIND dataset: batch=4, lr=2e-5, epoch=2, coefficient=1.0
Arxiv dataset: batch=4, lr=2e-5, epoch=2, coefficient=0.8
NSF dataset: batch=4, lr=2e-5, epoch=2, coefficient=0.4
bash run.sh
@inproceedings{cui2020bert,
title={BERT-enhanced relational sentence ordering network},
author={Cui, Baiyun and Li, Yingming and Zhang, Zhongfei},
booktitle={Proceedings of EMNLP},
pages={6310--6320},
year={2020}
}
Some codes refer to the paper “Enhancing Pointer Network for Sentence Ordering with Pairwise Ordering Predictions” and "Fine-tune BERT for Extractive Summarization". Thanks!