Skip to content

thu-spmi/LABES

Repository files navigation

LABES

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]

Requirements

  • 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

Data Preparation

  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.

Running Experiments

Training

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

Testing

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

Reproducibility

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

Bug Report

Feel free to create an issue or send email to zhangyic@umich.edu

About

Yichi Zhang et al. A Probabilistic End-To-End Task-Oriented Dialog Model with Latent Belief States towards Semi-Supervised Learning. EMNLP 2020.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages