NLP Sentiment analysis for abuse, spam and malicious bot detection is something I would like to consider in the long term.

It could be trained using actionable reported content like removing a spam post or comment! It would be an optional feature and fun to work on 😁