thu-spmi / PARG

Paraphrase Augmented Task-Oriented Dialog Generation, ACL 2020

Source Code for the ACL 2020 paper "Paraphrase Augmented Task-Oriented Dialog Generation"

Construct Paraphrase for CamRest676

python analysis.py

Construct Paraphrase for MultiWOZ

python analysis_multi.py

Training with default parameters

python model.py -mode train -model [tsdf-camrest|tsdf-multiwoz]

(optional: configuring hyperparameters with cmdline)

python model.py -mode train -model [tsdf-camrest|tsdf-multiwoz] -cfg lr=0.003 batch_size=32

Testing

python model.py -mode test -model [tsdf-camrest|tsdf-multiwoz]

Reinforcement fine-tuning

python model.py -mode rl -model [tsdf-camrest|tsdf-multiwoz] -cfg lr=0.0001

Before running

1. Install required python packages. We used pytorch 0.3.0 and python 3.6 under Linux operating system.

pip install -r requirements.txt

2. Download CamRest676 dataset and MultiWOZ dataset and place them in PROJECT_ROOT/data/CamRest676 and PROJECT_ROOT/data/MultiWOZ, respectively.

3. Make directories under PROJECT_ROOT.

mkdir vocab
mkdir log
mkdir results
mkdir models
mkdir sheets

4. Download pretrained Glove word vectors and place them in PROJECT_ROOT/data/glove.