NLP classification problem?

Hi all, can you help me think through this problem? I’m a college teacher and I have an article analysis assignment. I’m creating a chatbot where students can get a preliminary evaluation of the assignment before they turn it in. The assignment is about 500 words. Part of the assignment is to include a discussion of what was left out. The problem is there is a wide enough variety of ways of talking about what was left out that I can’t just look for a set of keywords. This feels to me like an nlp classification problem. I have about 80 pre-labeled examples. But there are two considerations that I’m not quite sure how to deal with:

First, the classification is present or absent, but I’m not sure how to train a model for “not there,” is there a term for this kind of problem?

Second, the assignments are large enough I’m worried given the small number of examples the model will not be focused on what I want. I suppose the obvious answer is to cut up the examples so that it only has data I’m looking for. Is the easier answer?
Thanks in advance.