“We already store data. In a database. It works well”

Re:“We already store data. In a database. It works well” by raghavtoshniwal on Tuesday February 11, 2020 @07:57AM (#22298076)

Off topic but it is considered difficult for NLP models to do sarcasm detection (because it’s a really hard problem, no widely available datasets, some context awareness is required, etc.)

For better or worse, in the last few years we have been generating a dataset on social forums by using “/s” at the end of sarcastic statements.

I’ll try and mine the “/s” statements on HN at least and see if I get somewhere.

[ Parent | on HN ]

Re:“We already store data. In a database. It works well” by skrebbel on Tuesday February 11, 2020 @10:27AM (#22299174)

That's going to be a great data set! Most people who use "/s" are great at sarcasm.

[ Parent | on HN ]

Re:“We already store data. In a database. It works well” by tveita on Tuesday February 11, 2020 @10:01AM (#22298955)

There is some prior art for this https://arxiv.org/abs/1704.05579

[ Parent | on HN ]

Re:“We already store data. In a database. It works well” by duckmysick on Tuesday February 11, 2020 @09:18AM (#22298626)

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.

[ Parent | on HN ]