A mentee of mine recently asked me this question as he prepares his manuscript on a novel machine learning method in human computer interaction. I suspect this is a common question for folks who are starting to write research papers for publication, especially in STEM fields. Unlike other sections such as Experiments or Results, the Discussion section can be confusing as to what content should go in. Below I list a few questions an author could ask themself — and when you have answers to these questions ready, you will have the right content to populate a nice Discussion section already.
1. Do any parts of your results seem surprising/unexpected/particularly interesting to you, good or bad? Why might that have happened?
2. Do any parts of your results agree with or contradict domain knowledge and theory? Discuss the domain knowledge and theory and how they agree or contradict. (This is where a collaborative, domain knowledge expert can chime in.)
3. If your proposed methods worked in your experiments, under what circumstances (data, participants, technology, etc.) will it be reproducible or not reproducible in another study?
4. If your proposed methods worked in your experiments, what other applications/tasks may your methods also be good (or actually be bad) at?
This is by no means an exhaustive list but it should be a very good start to compose a strong Discussion section for your next paper.