This is the official implementation for the paper "BIG-FUSION: Brain-Inspired Global-Local Context Fusion Framework for Multimodal Emotion Recognition in Conversations".

• Python 3.9.19
• PyTorch toolbox (2.1.0+cu121)
• Torch-geometric 2.4.0

Please use the same CUDA version for training.

We use the data from paper "A Transformer-based Model with Self-distillation for Multimodal Emotion Recognition in Conversations".

To preprocess the data, please extract multimodal features as described in the paper and use the following preprocessing command:

python preprocess.py --dataset="IEMOCAP"

python train.py --dataset="iemocap" --modalities="atv" --from_begin --epochs=60

python eval.py --dataset="iemocap" --modalities="atv"

If this repository is useful for you, please cite as:

@inproceedings{Wang2025BIGFUSIONBG,
  title={BIG-FUSION: Brain-Inspired Global-Local Context Fusion Framework for Multimodal Emotion Recognition in Conversations},
  author={Yusong Wang and Xuanye Fang and Huifeng Yin and Dongyuan Li and Guoqi Li and Qi Xu and Yi Xu and Shuai Zhong and Mingkun Xu},
  booktitle={AAAI Conference on Artificial Intelligence},
  year={2025},
}