Yet another Transformer model fine-tuned for approximating another non-linear mapping between X and Y? That's right! This is your good ol' emotion classifier - given an input text, the model outputs a probability distribution over a set of pre-selected emotion words. In this case, it is 32, which is the number of emotion classes in the Empathetic Dialogues dataset.
This model is built "on top of" a distilbert-base-uncased model fine-tuned on the go-emotions dataset. Y'all should really check out that model, it even contains a jupyter notebook file that illustrates how the model was trained (bhadresh-savani if you see this, thank you!).
Well where should we begin...
- Unable to ascertain the degree of cultural specificity for the context that a respondent described when given an emotion label (i.e., p(description | emotion, culture))
- Downloads last month