Humans have problems detecting written sarcasm. Not surprising that NLP models struggle with it too. It's difficult because even if you have the context, the o

Humans have problems detecting written sarcasm. Not surprising that NLP models struggle with it too. It's difficult because even if you have the context, the original poster can change classification after being called out by replying "oh, I was just kidding/I was being sarcastic". The same comment may and may not be sarcastic at the same time, Schrödinger's cat style. That being said, it's an interesting problem. Hope you find something good.