code for '"SUNET: Speaker-utterance interaction Graph Neural Network for Emotion Recognition in Conversations' published by 'Engineering Applications of Artificial Intelligence'.
We used the Roberta Embedding offered by DAG-ERC. Of course, you can get better results by training your Roberta.
When you donwload the Roberta Embedding or train yourself, process.py can be run to get speakers' feastures.
Then, tran.py is executed to train the model.
config.py is used to select different data sets and corresponding parameters.