Skip to content

ZJLAB-AMMI/msp

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

This project contains the source code and the dataset to reproduce the main result of the paper "On Tracking Dialogue State by Inheriting Slot Values in Mentioned Slot Pools".

We have tested the validity of this project at a machine equipped with Ubuntu 18.04.3 LTS system, an Intel Xeon Silver 4214 CPU, 256 GB main memory, and 8 Nvidia Tesla V100 video cards.

This paper has been accepted as a poster paper of IJCAI-ECAI 2022.

Requirements

Please create a virtual python environment (or conda environment), and then install all packages listed in the "requirements.txt" into the environment.

The training phase of this model (in the default setting) requires about 14GB video memory. If your available video memory is less than 14GB, please decrease the batch size to an appropriate value (by using the command "python train.py --batch_size=*").

Execution

  1. Uncompress the multiwoz21.zip file.

    The directory should be like:

   msp/
     train.py
     multiwoz21.zip
     ...
     multiwoz21/
       data.json
       ...
  1. Activate the environment you just created.
  2. Run the "train.py" script to train and test the model. The result will be saved in the "/evaluation" folder that automatically created by the script.

Note

  1. The data preprocess procedure and the pretrained model download procedure may take a long time when you run the script first time, please wait patiently.
  2. You can set the "use_multi_gpu" to True (by using the command "python train.py --multi_gpu=True") to accelerate the training speed if you have multiple video cards.
  3. We did not upload the MultiWOZ 2.2 dataset into the project to save space. Please follow the instructions below if you want to train and test the model via the MultiWOZ 2.2 dataset.
    1. Download the MultiWoZ 2.2 dataset from the project.
    2. Follow the instruction in the project , converting the MultiWOZ 2.2 dataset into the MultiWOZ 2.1 dataset format.
    3. Save the restructured dataset into a folder named "multiwoz22" under the "msp" directory.
    4. Modify the value of the 9th line of config.py from "multiwoz21" to "multiwoz22".
    5. run the "train.py" script.

Citation

Please cite our paper if you think our study is helpful to your research.

@article{sun2022tracking,
  title={On Tracking Dialogue State by Inheriting Slot Values in Mentioned Slot Pools},
  author={Sun, Zhoujian and Huang, Zhengxing and Ding, Nai},
  journal={arXiv preprint arXiv:2202.07156},
  year={2022}
}

About

Source code of mentioned slot pool

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages