This repo contains the implementation of the paper "Consistent Dialogue Generation with Self-supervised Feature Learning"
Tensorflow == 1.5.1
Training CoCon Discriminator
python feature_extractor.py for feature extraction.
Training CoCon Generator
python cocon.py -global # using global/topic feature to training a CoCon-T
python cocon.py -global -local # using both global/topic and local/persona feature to training a CoCon-TP
Other option information can be found in the help python cocon.py --help.
Generating controlled response by modifying bits:
python interpolation.py -global -local --feed -tf /newdata2/test_old.txt --bit 25
Our paper can be cited at
@article{zhang2019consistent,
title={Consistent Dialogue Generation with Self-supervised Feature Learning},
author={Zhang, Yizhe and Gao, Xiang and Lee, Sungjin and Brockett, Chris and Galley, Michel and Gao, Jianfeng and Dolan, Bill},
journal={arXiv preprint arXiv:1903.05759},
year={2019}
}