HGNN

Code for Infusing Multi-Source Knowledge with Heterogeneous Graph Neural Network for Emotional Conversation Generation (AAAI21)

Introduction

The implementation is based on ReCoSa. Download MELD file and DailyDialog.

Requirement:

• nltk>=3.2.4

• numpy>=1.13.0

• regex>=2017.6.7

• tensorflow=1.10.1

• Scipy=1.1.0

Generate data

• For MELD

Generate the data format and the matrix A for training (The uploaded train.txt, dev.txt, and test.txt are the facial features extracted by Openface. You can also use the openface4extract_pic_feature.py code to generate by yourself.)

python generate_data4meld.py train/test/dev

python generate_matrix_A4meld.py train/test/dev

python generate_speakers4meld.py (generate speakers' name)

• For DailyDialog

You need to modify the code to remove the speaker, the image part to train on Dailydialog.

python generate_data_matrix_A_4dailydialog.py

Training

1、parameter setting:

hyperparams.py

2、To generate vocab:

python prepro.py

3、To train:

python train.py

4、To eval:

python eval.py

5、The raw dialogue data：

joey: or ! or , we could go to the bank , close our accounts and cut them off at the source .       neutral

chandler: you 're a genius !    joy

joey: aww , man , now we wo n't be bank buddies !   sadness

chandler: now , there 's two reasons .  neutral

Souce looks like:

or ! or , we could go to the bank , close our accounts and cut them off at the source .     neutral  </d>  

or ! or , we could go to the bank , close our accounts and cut them off at the source .     neutral  </d>  you 're a genius !       joy  </d>  

or ! or , we could go to the bank , close our accounts and cut them off at the source .     neutral  </d>  you 're a genius !       joy  </d>  aww , man , now we wo n't be bank buddies !      sadness  </d>  

Target:

chandler  </d>  you 're a genius !  </d>  joy  </d> 

joey  </d>  aww , man , now we wo n't be bank buddies !  </d>  sadness  </d>  

chandler  </d>  now , there 's two reasons .  </d>  neutral  </d>  

Citation

If you find this project helps, please cite our paper :)

@article{Liang_Meng_Zhang_Chen_Xu_Zhou_2021, 
    title={Infusing Multi-Source Knowledge with Heterogeneous Graph Neural Network for Emotional Conversation Generation}, 
    volume={35}, 
    url={https://ojs.aaai.org/index.php/AAAI/article/view/17575}, 
    number={15}, 
    journal={Proceedings of the AAAI Conference on Artificial Intelligence}, 
    author={Liang, Yunlong and Meng, Fandong and Zhang, Ying and Chen, Yufeng and Xu, Jinan and Zhou, Jie}, 
    year={2021}, 
    month={May}, 
    pages={13343-13352} 
}