Computer science researchers at the University of Central Florida, including one of Indian ancestry, have created an artificial intelligence (AI)-based sarcasm detector for social media messages.
Sarcasm has been a major impediment to improving sentiment analysis accuracy, especially on social media, since sarcasm depends heavily on vocal tones, facial expressions, and movements that cannot be expressed in text.
If AI refers to rational data processing and reaction, sentiment analysis is similar to accurately recognizing emotional expression on social media.
“The inclusion of sarcasm in the text is the greatest hindrance in the success of emotion analysis,” says Ivan Garibay, Assistant Professor of engineering at the University of Central Florida’s Complex Adaptive Systems Lab (CASL).
Sarcasm isn’t often easy to spot in speech, so you can imagine how difficult it would be for a computer program to do it well.
“We created an interpretable deep learning model using multi-head self-attention and gated recurrent units,” Garibay wrote in a paper published in Entropy.
The researchers trained the computer model to recognize patterns that often signify sarcasm, and then paired that with training the algorithm to accurately identify cue words in sequences that were more likely to indicate sarcasm.
They trained the model to do this by feeding it vast amounts of data and then testing its accuracy.
Ramya Akula, a doctoral student in computer science, was part of the project.
“In a face-to-face conversation, sarcasm can be easily detected using facial expressions, movements, and the speaker’s tone,” Akula said.
“Because none of these cues are readily accessible, detecting sarcasm in textual speech is a difficult challenge. Sarcasm identification in online conversations from social networking sites has become much more difficult, particularly with the explosion in internet use “She continued.