The US Secret Service is looking for a way to detect sarcasm on Twitter, posting a tender notice online to find coders capable of creating the software.
The agency said they were looking for a program with the “ability to detect sarcasm and false positives” along with “sentiment analysis” and “influencer identification”. They also requested “access to historical Twitter data”.
Ed Donovan, a spokesperson for the Secret Service, told Sky: “Our objective is to automate our social media monitoring process. The ability to detect sarcasm and false positives is just one of 16 or 18 things we are looking at.”
The contract offered by the agency covers a five-year period, with the tender noting that the program must also be compatible with Internet Explorer 8 – a piece of software released back in 2006 and for which free upgrades are available.
Despite the obvious benefits of government agents being able to sort credible threats from the merely facetious, recognizing sarcasm – or most nuanced moods – has so far proved difficult for computers.
A study from 2011 conducted by the University of New Jersey collected tweets with the hashtags ‘sarcasm’ or ‘sarcastic’ and then removed these markers, mixing the messages in with others that displayed “positive or negative attitudes without sarcasm”.
The researchers found that at best, software programs could only sort the sarcastic from the sincere around 65 per cent of the time, although human volunteers didn’t do much better, achieving accuracies of between 62 and 72 per cent with the same material.
It seems that without the many contextual clues of what you know about an individual’s personality or can be gleamed from face to face interactions both humans and machines will have to struggle with sarcasm online. Which is great.