With so much research going on sarcasm or sentiment analysis on twitter, is there any state-of-the-art or benchmark method for sarcasm detection, to which performance of new methods can be compared; may be on a standard data-set of sarcastic tweets?
Anybody used any such approach to compare their method or implementation ?
Interesting question
There is this paper: Automatic Sarcasm Detection: A Survey. he treats the approaches on short texts and long texts
Article Automatic Sarcasm Detection: A Survey
I would be interested by a dataset too.
The above-mentioned paper doesn't mention DeepMoji which is, if I remember correctly, the best sarcasm recognizer so far.
https://deepmoji.mit.edu
You have a perfect opportunity to create such a standard/benchmark by creating a lexicon in a bottom-up approach. It may not be easy though.
I will take that as a NO to my question.
Sure @Meliha . But some collaboration will be required.
There may be linguistic markers, but essentially the detection of sarcasm depends on recognizing the speech event and having a social concept of the perspective of the speaker. Sarcasm is a replica of something sincerely intended, only in the wrong context or by a speaker known to think opposite of the statement. Sometimes sarcasm is marked by an opening expression, like "right" or "sure", "of course" to mimic quoting someone. You might want to check whether those marker words appear in whatever texts you are searching, but you will still have to assess whether the comment is something the speaker, writer actually believes based on their other positions. Sometimes people have to inform others who take the statement literally that in fact they were being sarcastic.
- Badges
- Science topic