Sentiment analysis of movie reviews is not what i would call an inspired or high-quality programming problem. This seems like too common examples from the trai

Sentiment analysis of movie reviews is not what i would call an inspired or high-quality programming problem. This seems like too common examples from the training data being made manifest. There's so many more interesting things you could be doing -- Even if you stick with the "use tf-idf on review data" how about sentiment analysis of vacation destination reviews segmented by season posted? Things like that lead directly into other ideas and possible metrics. Creativity is really a sore spot with these. I suspect more elaborate prompts can suppress the commonalities but gpt-3.5 with a bog standard prompt gives bog standard ideas.