I want to fine tune BERT to perform multi-class multi-label classification. Input to the model is review. Output should be the brand name, the sentiment associated, the note (Price, Interior design, exterior design, comfort,durability etc), the type(luggage capacity, charging, front view , rare view,colour, dashboard etc) and subtype (Black, charging time, seat capacity etc).
If a review has more than one thing. The model should duplicate the review to as many times and predict the set of labels(Brand name, Sentiment, Notes, Type and Sub Type)
I have built a model which classifies brand name, note, type subtype and sentiment. But the issue is, it is not producing multiple outputs for a single input.
Basically, I want know how to fine tune a BERT to produce multiple output rows for a single input row