This is the code for EMNLP 2020 paper "A Probabilistic End-To-End Task-Oriented Dialog Model with Latent Belief States towards Semi-Supervised Learning". [paper link]

• Python 3.6
• PyTorch 1.2.0
• NLTK 3.4.5

We use some NLP tools in NLTK which can be installed through:

python -m nltk.downloader stopwords punkt wordnet

1. Unzip raw data of CamRest676, Stanford In-Car Assistant and MultiWOZ 2.1, and also the GloVe word embeddings into the corresponding directories. Note that file "compressed_data_2.0.json.zip" is the raw MultiWOZ 2.0 data from this repository, for normalizing entity names in the data preprocessing process.

2. Data Preprocess Raw data are preprocessed automatically during the first run of each dataset. See datasets.py and multiwoz_preprocess.py for what have been done in the data preprocessing process.

python train.py -mode train -dataset [camrest|kvret|multiwoz] -method cvae -c spv_proportion=[a integer between 0-100] exp_no=your_exp_name

python train.py -mode test -dataset [camrest|kvret|multiwoz] -method cvae -c eval_load_path=[experimental path]

We release the models that obtain the best results in Table 1 and Table 2. Run the following commands for model evaluation.

python train.py -mode test -dataset camrest -method cvae -c eval_load_path=experiments/camrest/camrest_best beam_search=True
python train.py -mode test -dataset kvret -method cvae -c eval_load_path=experiments/kvret/kvret_best beam_search=True
python train.py -mode test -dataset multiwoz -method bssmc -c eval_load_path=experiments/multiwoz/multiwoz_best beam_search=True

Feel free to create an issue or send email to [email protected]