rfm's Introduction

RFM: response-aware feedback mechanism for background based conversation

The code for the paper RFM: response-aware feedback mechanism for background based conversation.

Reference

If you use any source code included in this repo in your work, please cite the following paper.

@article{chen2022rfm,
  title={RFM: response-aware feedback mechanism for background based conversation},
  author={Chen, Jiatao and Zeng, Biqing and Du, Zhibin and Deng, Huimin and Xu, Mayi and Gan, Zibang and Ding, Meirong},
  journal={Applied Intelligence},
  year={2022},
  publisher={Springer},
  doi={10.1007/s10489-022-04056-4}
}

Requirements

python 3.7
pytorch 1.7.0

Datasets

Download the raw data version of Holl-E, and put the raw data files (train_data.json, dev_data.json and test_data.json) in the directory dataset/holl/raw_data/.
Then, run the preprocessing script:

python Prepare_holl.py

As for Wizard of Wikipedia(WoW) dataset, we use the DukeNet version. Download the Wizard of Wikipedia dataset, and put data files in the directory dataset/wizard_of_wikipedia/.
Download the glove.6B.300d.txt and put it in dataset/holl_e/oracle/, dataset/holl_e/mixed/ and dataset/wizard_of_wikipedia/.

Run training, validation, and testing

To train or test your model, run:

# Holl-E dataset
python -m torch.distributed.launch --nproc_per_node=num_GPU Run_RFM_Holl.py --mode='train/test'

# Wizard of Wikipedia dataset
python -m torch.distributed.launch --nproc_per_node=num_GPU Run_RFM_WoW.py --mode='train/test'

If you want to run multiple references(MR) test version in Holl-E dataset, please add --test='MR' in the run script.