Source code for our paper "Improving Empathetic Response Generation by Recognizing Emotion Cause in Conversations"

Source code for our paper "Improving Empathetic Response Generation by Recognizing Emotion Cause in Conversations"

this repository is maintained by both Jun Gao and Yuhan Liu

Environment Requirment

• pytorch >= 1.4.0
• texar.torch
• bert-score
• nltk

Model Overview

Running

1. we use RECCON to train an emotion cause detection model and apply it to annatate EmpatheticDialogues. The processed data is in Data.

2. Then you need to pretrain the emotion classification model, here you need to download glove.6B.300d first and then running the following command. Here $GLOVE is the glove embedding file:

   bash ./bash/run_emotion.sh --glove $GLOVE --gpu_id 0

3. To train the model and generate the automatic metric results, firstly you need to make sure that bert-score is successfully installed. In our paper, we use roberta-large-en rescaled with baseline to calculate BERTScore. You can download roberta-large-en from Hugginface. For the rescaled_baseline file, we can download it from here.

   Then run the following command. Here $ROBERTA_DIR is the downloaded roberta-large-en model directory and $BASELINE is downloaded baseline file.

   to train soft-gate model:

   bash ./bash/run_generation.sh --glove $GLOVE --gpu_id 0 --mode soft --roberta $ROERBTA_DIR --baseline $BASELINE --do_train

   to test soft-gate model:

   bash ./bash/run_generation.sh --glove $GLOVE --gpu_id 0 --mode soft --roberta $ROERBTA_DIR --baseline $BASELINE --do_test

   to train hard-gate model:

   bash ./bash/run_generation.sh --glove $GLOVE --gpu_id 0 --mode hard --roberta $ROERBTA_DIR --baseline $BASELINE --do_train

   to test hard-gate model:

   bash ./bash/run_generation.sh --glove $GLOVE --gpu_id 0 --mode hard --roberta $ROERBTA_DIR --baseline $BASELINE --do_test

Acknowledgement

@inproceedings{gao-etal-2021-improving-empathetic,
    title = "Improving Empathetic Response Generation by Recognizing Emotion Cause in Conversations",
    author = "Gao, Jun  and Liu, Yuhan  and Deng, Haolin  and Wang, Wei  and Cao, Yu  and Du, Jiachen  and Xu, Ruifeng",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2021",
    month = nov,
    year = "2021",
    pages = "807--819",
    publisher = "Association for Computational Linguistics"
}