Unsupervised Dialog Structure Learning

Codebase for Unsupervised Dialog Structure Learning, published as a long paper in NAACL 2019. The codebase is developed based on NeuralDialog-CVAE.

If you use the datasets or any source codes included in this repository in your work, please cite the following paper. The bibtex is listed below:

@article{shi2019unsupervised,
 title={Unsupervised Dialog Structure Learning},
 author={Shi, Weiyan and Zhao, Tiancheng and Yu, Zhou},
 journal={arXiv preprint arXiv:1904.03736},
 year={2019}
}

pytorch implementation with structured attention

https://github.com/Liang-Qiu/SVRNN-dialogues

Requirements

Listed in the requirements.txt. I was using a pretty old tensorflow version when first developing the project. I think most of the major functions are still supported but haven't tested it on the new versions yet.

python 2
tensorflow == 1.0.1

Train:

python main.py --result_path data/results/whatever_name.pkl

Test:

python main.py --result_path data/results/whatever_name.pkl --forward_only True --test_path runSomeTimeStamp

After the training, there will be a directory at working/runSomeTimeStamp (e.g. run1532935232), just copy the dir name and pass it to test_path.

Interpretation

In interpretation.py.

Datasets

1. CamRest676
2. SimDial