Recently, US researchers unveiled a sarcasm detector for social media. Perhaps one day, this kind of work could be at the origin of a real revolution concerning the moderation strategy of the different platforms.
Identify sarcastic cue words
Sarcasm is a ironic mockery that can sometimes be seen as a form of “prickly” or belligerent irony. Unlike the irony that involves saying the opposite of what you think by making it understood that you think the opposite of what you say, sarcasm goes a little further. Indeed, it is a question of saying the opposite of what one thinks, without showing that one thinks the opposite of what one says. Thus, when a person makes sarcastic remarks, it is not not always obvious to detect it. Face to face, it is possible to observe certain gestures or facial expressions that may possibly betray the interlocutor in question. On the other hand, when it comes to a comment on social media, spotting the sarcasm becomes much more difficult.
In a study published in the journal Entropy on March 26, 2021, a duo of researchers from the University of Central Florida (United States) explained that they had developed their own sarcasm detector. The latter is intended for use on social networks. Researchers said they used deep learning to develop a system able to identify “key words” sarcastic and analyze the dependencies between these words in order to qualify (or not) the comment as sarcastic.
A tool that can be of interest to social networks
Thus, the tool that the researchers developed can distinguish between ironic mockery and real hate speech. The authors of the study have not communicated their success rate, but according to them the results are rather convincing. Obviously, artificial intelligence still has some work to do to get to the human level (moderators) when it comes to spotting sarcasm online. However, if AI one day achieves a very high level of performance, there is no doubt that the major social networks will be interested in this type of research.
Researchers at the University of Central Florida are not the only ones who have done this kind of work. In November 2020, scientists from the Beijing Academy of Sciences (China) have developed a system based on data from Twitter. The AI in question takes into account the hashtags present in the comments in order to better understand the state of mind of the user. However, this model would have a 83% success rate!