A new artificial intelligence system is able to recognise when someone is being sarcastic online – something even humans sometimes struggle to do.
Social media chatter is mined by advertisers, stock traders and intelligence services, but their systems can be thrown by sarcasm.
Because computers process information literally, they struggle to tell when someone is actually saying the opposite of what they really mean.
In 2014, the US Secret Service listed sarcasm detection as a feature it wanted implemented in its social media analytics service.
Now Silvio Amir from the University of Lisbon has come up with a machine learning technique to overcome this problem on Twitter.
He said sarcasm can be identified by computers if they comb through the user’s previous tweets.
The system builds up a picture of the person, which gives it a better chance of deciding whether a person is being sarcastic or not.
He told the New Scientist: “It intuitively makes sense. Tell me what you talk about and I can tell you who you are.
“We can get away without looking at all this external information.”
He says the system has an accuracy level of about 87%, but that could improve as the system learns over time.
The system could be adapted to work with any platform where previous posts are available for analysis.