zhuo-zhi / Improving-NQG-with-CGAN

Code about Improving Nerual Question Generation with adversarial training

Improving Neural Question Generation with CGAN

This repository contains code about Improving NQG with CGAN, which adds a discriminator to do adversarial training to improve the performance of neural question generation network.

(image from the reference below)

About the Code

Environment: Pytorch 4.0, Python 3.6

Package Requirement: torchtext, nltk, numpy

Run on GPU: Python train.py

Performance

Pre-trained MLE Model: BLEU-4: 8.3

Model after adversarial training: BLEU-4: 8.5

Although the BLEU-4 metric is not very high in the experiment, this code provides a basic framework which might be useful.

Reference

• Improving Neural Machine Translation with Conditional Sequence Generative Adversarial Nets