EmpDG: Multi-resolution Interactive Empathetic Dialogue Generation (COLING 2020)

This repository is the implementation of our COLING 2020 Paper EmpDG: Multi-resolution Interactive Empathetic Dialogue Generation.

Model Architecture

Setup

• Check the packages needed or simply run the command:

pip install -r requirements.txt

• The dataset (empathetic-dialogue) is preprocessed and stored in pickle format:

.
└── empathetic-dialogue
    └── empdg_dataset_preproc.p

• Download GloVe vectors from here (glove.6B.300d.txt) and put it into /vectors/.

• For reproducibility purposes, we place the model checkpoints at Google Drive. You could download and move it under /result/.

• To skip training, please check folder /Predictions/.

Run code

Training

EmpDG

# 1. Pre-train Empathetic Generator (EmpDG_woD)
# 2. Pre-train two Interactive Discriminators
# 3. Joint-train two components
python3 adver_train.py --cuda --label_smoothing --noam --emb_dim 300 --rnn_hidden_dim 300 --hidden_dim 300  --hop 1 --heads 2 --emotion_disc --pretrain_emb --model EmpDG --device_id 0 --save_path save/EmpDG/ --d_steps 1 --g_steps 5 --pointer_gen

EmpDG_woD

# Empathetic Generator
python train.py --cuda --label_smoothing --noam --emb_dim 300 --hidden_dim 300 --hop 1 --heads 2 --pretrain_emb --model EmpDG_woD --device_id 0 --save_path save/EmpDG_woD/ --pointer_gen

EmpDG_woG

# We regard the baseline EmoPrepend as generator, which only considers the coarse-grained emotional factor. 
# We only use the semantic discriminator to distinguish the generated responses and the gold ones. 

python3 adver_train_no_eg.py --cuda --label_smoothing --noam --emb_dim 300 --rnn_hidden_dim 300  --hidden_dim 300 --hop 1 --heads 2 --cuda --pretrain_emb --model EmpDG_woG --device_id 0 --save_path save/EmpDG_woG/ --d_steps 1 --g_steps 5 --pointer_gen 

Add --resume_g --resume_d into the command to directly use the pre-trained generator or discriminator.

Testing

EmpDG

Make sure that the trained model result/EmpDG_best.tar exists.

python3 train.py --test --model EmpDG --cuda --label_smoothing --noam --emb_dim 300 --rnn_hidden_dim 300 --hidden_dim 300  --hop 1 --heads 2 --pretrain_emb --device_id 0 --save_path save/EmpDG/ --pointer_gen

Interaction

python3 interact.py --model EmpDG --cuda --label_smoothing --noam --emb_dim 300 --rnn_hidden_dim 300 --hidden_dim 300  --hop 1 --heads 2 --pretrain_emb --device_id 0 --save_path save/EmpDG/ --pointer_gen

• Generation examples from an interactive environment:

Role	Response
User	i am so excited because i am finally going to visit my old friend !
EmpDG	that is awesome ! what kind of visit ?
User	we will enjoy chatting and playing .
EmpDG	that is a lot of nostalgia .

Role	Response
User	i argued with my brother .
EmpDG	oh no ! what happened ?
User	i don't want to recall .
EmpDG	i am sorry , that is not a good thing .

Reference & Acknowledgements

If you find our work useful, please cite our paper as follows:

@inproceedings{li2020empdg,
  title={EmpDG: Multi-resolution Interactive Empathetic Dialogue Generation},
  author={Li, Qintong and Chen, Hongshen and Ren, Zhaochun and Ren, Pengjie and Tu, Zhaopeng and Chen, Zhumin},
  booktitle={Proceedings of the 28th International Conference on Computational Linguistics},
  pages={4454--4466},
  year={2020}
}