├── checkpoints                     # dir for saving best model and tensorboard logs.
├── logs                            # dir for recording outputs of the training process.
├── pretrained_models               # dir for saving pretrained models from HuggingFace.
├── results                         # dir for recording performance of the SKEAFN model.
├── ca.py                           # implement for text-guided interaction.
├── classifier.py                   # implement for classifier.
├── dataloader.py                   # implement for constructing PyTorch dataloader.
├── fafw.py                         # implement for feature-wised attention fusion.
├── get_data.py                     # implement for getting train/valid/test dataloader.
├── logger.py                       # implement for obtaining tensorboard log.
├── loss.py                         # implement for ASLSingleLabel loss.
├── main.py                         # implement for main.
├── metrics.py                      # implement for evaluation metrics.
├── model.py                        # implement for our SKEAFN model.
├── tools.py                        # implement for train tools.
├── train.py                        # implement for the train process.
├── trick.py                        # implement for tricks in training process.
├── LICENSE                     
└── README.md

pip install -r requirements.txt

nohup python main.py --dataset_name=twitter --category_number=2 --pretrained_arch=roberta --num_workers=2 > logs/twitter_roberta_train.log 2>&1 &

nohup python main.py --dataset_name=twitter --test_pretrained_path=best_model.pth > logs/twitter_roberta_test.log 2>&1 &

@article{ZHU2023101958,
  title = {SKEAFN: Sentiment Knowledge Enhanced Attention Fusion Network for multimodal sentiment analysis},
  journal = {Information Fusion},
  volume = {100},
  pages = {101958},
  year = {2023},
  issn = {1566-2535},
  doi = {https://doi.org/10.1016/j.inffus.2023.101958},
  url = {https://www.sciencedirect.com/science/article/pii/S1566253523002749},
  author = {Chuanbo Zhu and Min Chen and Sheng Zhang and Chao Sun and Han Liang and Yifan Liu and Jincai Chen},
  keywords = {Multi-view learning, Multiple feature fusion, Multimodal sentiment analysis, External knowledge, Multi-head attention},
}