we use sbert bi encoder and RoBERTa to implement model that learns from the context of dialogues & predicts emotion of all the users better.
left: model without context analysis
middle: with context analysis, without training on pseudo labelled data
right: with context analysis, training on pseudo labelled data