It is not always easy to tell when someone is making sarcastic remarks. However, face-to-face, certain cues, such as gestures or facial expressions, can serve as indicators. However, in a comment on social networks, it is not easy to see clearly.
It is this daunting challenge that researchers at the University of Central Florida have risen to by developing their own sarcasm detector for texts posted on the platforms. Specifically, the scientists provided their computer model with keywords and sequences that they believed were most likely to contain ironic statements.
Another system achieves 86% success
The tool can therefore distinguish between real hate speech and a simple joke between two online friends. We don’t have an exact success rate for this artificial intelligence, but the results would be convincing enough.
Over time, algorithms are increasingly able to detect sarcasm, even if its level remains lower than that of a human moderator. In any case, we can imagine that the major social networks will have a lot of interest in all of this work.
These researchers are not the first to work on this main topic. We talked to you in particular last November about this AI that can detect sarcastic remarks. Chinese scientists developed this system using data from Twitter. In particular, it relies on the hashtags that are present, which make it possible to better assess the state of mind of the Internet user and the tone he wants to give his message.
If the text turns out to be a topic and the picture says the opposite, the tool will think it’s irony. The results seem to be very good again and the detection model achieved 86% compared to 83% for another system called HFM, which was introduced in 2019.