mkfm's Introduction

MKFM

The official implementation for Findings of the EMNLP 2023 paper An Empirical Study on Multiple Knowledge from ChatGPT for Emotion Recognition in Conversations.

Requirements

Python 3.7.11
PyTorch 1.8.0
Transformers 4.1.1
CUDA 11.1

Preparation

Download datasets and save them in ./data.

Download knowledge and save them in ./.

Training & Evaluation

You can train the models with the following codes:

--TP : Using auxiliary label knowledge: topic
--SC : Using auxiliary label knowledge: sarcasm
--MP : Using auxiliary label knowledge: metaphor
--EC : Using auxiliary utterance knowledge: emotional cause
--CS : Using auxiliary utterance knowledge: commonsense knowledge
--ACS : Using auxiliary utterance knowledge: affective commonsense knowledge
--CR : Using auxiliary contextual knowledge: co-reference
--CT : Using auxiliary contextual knowledge: context
--EC2 : Using auxiliary contextual knowledge: emotional cause

For IEMOCAP: python run.py --dataset IEMOCAP --gnn_layers 4 --lr 0.0005 --batch_size 16 --epochs 30 --dropout 0.2

For MELD: python run.py --dataset MELD --lr 0.00001 --batch_size 64 --epochs 70 --dropout 0.1

For EmoryNLP: python run.py --dataset EmoryNLP --lr 0.00005 --batch_size 32 --epochs 100 --dropout 0.3

Citation

If you find our work useful for your research, please kindly cite our paper as follows:

@inproceedings{tu2023empirical,
  title={An Empirical Study on Multiple Knowledge from ChatGPT for Emotion Recognition in Conversations},
  author={Tu, Geng and Liang, Bin and Qin, Bing and Wong, Kam-Fai and Xu, Ruifeng},
  booktitle={Findings of the Association for Computational Linguistics: EMNLP 2023},
  pages={12160--12173},
  year={2023}
}