Can I use sentiment analysis to be more specific than neg, neutral and pos?

Question

I\'m writing a machine learning model that will translate a paragraph written by someone marking an exam paper into a grade (as a percentage). I\'ve got a dataset of over 3