On Monday this week the US Secret Service posted a tender which, among other things, solicits bids for social media analytics software that has the “ability to detect sarcasm”. It seems the spooks want to know whether you are being sincere when you tweet “God I just love having all my online activities monitored by the US intelligence services!” or if you’re cloaking your disdain using words that mean the exact opposite of what they say, like some kind of verbal ninja.
I’ll admit I’m struggling with the idea of how it will actually work. I took the opportunity to look through some recent sarky comments, particularly prevalent on Twitter during hashtag-hijacks like #WhyImVotingUKIP or #AskBG, and what became apparent is that sometimes I can’t tell if people are being serious or not. I could expand on what that means about UKIP voters in another blog, but what I mean here is that if even someone with a decent grasp of dry British humour can’t tell whether something’s sarcastic or not, how are you meant to construct an algorithm to find out?
One of the reasons this is difficult is because sarcasm is the backbone of social media. Most of what’s said online, at least in Britain and certainly on Twitter, is said with a tinge of verbal irony. I regularly see people visibly chuckle to themselves as they send their carefully-crafted 140-character missives. The ubiquity of sarcasm on social media explains why it’s so elusive to sentiment analysis software.
It’s important to remember that whether you’re monitoring for sarcasm, or just plain ‘positive’ and ‘negative’ comments, the principles of good sentiment analysis remain the same. The process will only be helpful if social insights specialists are asking the right questions. This point was touched on in an interesting webinar I listened in on last week, Understanding Context & Sentiment Analysis, hosted by Gorkana’s Lead Consultant for Social Media Insights, Alistair Wheate. Alistair’s main point was that just knowing whether messages are positive or negative, or in this case sarcastic, is not useful. Context is king, so asking good questions is essential.
Let me demonstrate the point. If the Secret Service wanted to use sarcasm-detecting software for their own Twitter presence, they might want to drill down into the data to find out: are people more or less sarcastic about the Secret Service than they are of, say, MI6? How sarcastic are people about intelligence services as a whole as compared to other sectors like telecoms companies? (“It’s just dandy that I don’t get signal in my central London home.”) How does this vary when you’re looking at comments about PRISM versus comments about career options? Asking the right questions makes big data valuable.
And of course ultimately, if you have to build software to understand the joke, the joke probably is on you.