RL for Question Generation
1.0.0
该存储库包含本文的代码和模型:探索特定问题的奖励,以产生深层问题(Coling 2020口头)。以下是我们提出的模型的框架。

pytorch 1.4.0
nltk 3.4.4
numpy 1.18.1
tqdm 4.32.2
scripts/train_example.sh以训练利用所有三个奖励的集成模型。 scripts/translate_example.sh以获取验证数据集上的预测。 我们使用MS COCO字幕生成的评估代码来评估自动指标。
pip install git+https://github.com/salaniz/pycocoevalcap
prediction.txt ,运行: python evaluate_metrics.py prediction.txt
@inproceedings{xie-etal-2020-RLQG,
title = {Exploring Question-Specific Rewards for Generating Deep Questions},
author = {Xie, Yuxi and Pan, Liangming and Wang, Dongzhe and Kan, Min-Yen and Feng, Yansong},
booktitle = {The 28th International Conference on Computational Linguistics (COLING 2020)},
year = {2020}
}