In different runtime environments, the results may be different. If there is any difference, please consider adjusting the weight of diss loss or the number of early stops.
You need to modify the PCA in the config, and the dataset and pretrained language model paths in the train file to match the paths on your device. If you switch to a different dataset, you also need to modify the corresponding parameters in both files.
Download the MOSI and MOSEI pkl file https://drive.google.com/drive/folders/1_u1Vt0_4g0RLoQbdslBwAdMslEdW1avI?usp=sharing. Put it under the "./dataset" directory.
Download the SentiLARE language model files https://drive.google.com/file/d/1onz0ds0CchBRFcSc_AkTLH_AZX_iNTjO/view?usp=share_link, and then put them into the "./pretrained-model/sentilare_model" directory. The setting "hidden_dropout_prob" in the config.json of the SentiLARE language model is set to 0.
If you do not want to train the model, you can load the pre-trained model weights. The training weights obtained on MOSI and MOSEI are available at the following link: https://drive.google.com/drive/folders/1T-ap6L6BhAZ8HbilUQI7GfM6Zd_ADVCP.
''' python train.py '''
Note: the scale of MOSI dataset is small, so the training process is not stable. To get results close to those in our paper, you can set the seed in args to 6758. The experimental results of this paper are obtained on the Linux system.