Description:
Right now, the model only outputs a label (useful/not-useful) and confidence level.
Goal:
We want an option that, when the PDF is classified as useful, also returns a few important keywords and the page numbers where they appear so users can quickly verify why the model chose that label.