MedDiaAMR

Code for the paper "Improving medical dialogue generation with abstract meaning representations"

Dataset and Resources

The dataset is from https://github.com/tangg555/meddialog

The structure of raw dataset should be like this:

├── dataset
       └── `train.source.txt`    
       └── `train.target.txt`       
       └── `val.source.txt` 
       └── `val.target.txt` 
       └── `test.source.txt` 
       └── `test.target.txt` 

To get the AMR graph (.concept and .path files) of each sentence, you can follow the preprocess in https://github.com/Soistesimmer/AMR-multiview

The structure of final dataset should be like this:

├── dataset
       └── `train.source.txt`
       └── `train.source.concept`
       └── `train.source.path`
       └── `train.target.txt`       
       └── `val.source.txt`
       └── `val.source.concept`
       └── `val.source.path`
       └── `val.target.txt` 
       └── `test.source.txt`
       └── `test.source.concept`
       └── `test.source.path`
       └── `test.target.txt` 

Preprocessing commands

bash preprocess.sh

Training commands

bash run-dual.sh

Citation

If you found this repository or paper is helpful to you, please cite our paper.

@misc{yang2023improving,
      title={Improving Medical Dialogue Generation with Abstract Meaning Representations}, 
      author={Bohao Yang and Chen Tang and Chenghua Lin},
      year={2023},
      eprint={2309.10608},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}