UK Edition
Oh, great. Robots can now tell when people are being sarcastic
A robot that can understand sarcasm in social media posts better than humans could provide a breakthrough in protecting people from abuse online.
Researchers at MIT have created an algorithm that can translate the tone of social media posts and understand if someone is being sarcastic.
The tool interprets emoji used in posts to help it understand emotional subtext and detect if sarcasm is being used.
Called DeepMoji, the AI analysed 1.2 billion tweets to understand how 64 popular emoji were used in them to convey meaning.
"Because we can't use intonation in our voice or body language to contextualise what we are saying, emoji are the way we do it online," said Iyad Rahwan, associate professor at the MIT Media lab, according to MIT Technology Review. "The neural network learned the connection between a certain kind of language and an emoji."
The researchers first taught the AI how to use appropriate emoji in given tweets based on whether the tone was sad, happy or funny. It then applied this emotional understanding to interpret the feeling behind messages.
The study is not the first to attempt to analyse sarcasm with a computer program, but the researchers said it is one of the most advanced to date. DeepMoji spotted sarcasm correctly 82 per cent of the time, outperforming human volunteers who had a 76 per cent accuracy rate.
It was initially designed as a tool that could detect racist posts on Twitter, but the researchers soon realised it needed to be able to translate sarcasm to be effective.
Understanding sarcasm in social media posts has been a key factor holding back monitoring tools for a number of years. Researchers have previously attempted to create computer programs that can understand the human phenomenon.
The US Secret Service said in 2014 that it was developing a way to detect sarcasm online as part of a broader plan to create its own social media monitoring tools.
In a separate study at the University of Lisbon last year, researchers developed a program that could understand if a tweet was sarcastic in tone by comparing it with a users' previous posts. The tool could spot sarcasm around 87 per cent of the time, according to the study.
Online abuse has become a scourge of social media companies, with the likes of Facebook, Twitter and YouTube struggling to police hateful material on their platforms. Technology giants have employed AI tools to help stem the phenomenon, training it to spot extremist videos and understand the context of posts.