The venerable US Secret Service Office of the Department of Homeland Security posted a work order this month. It is interested to acquire software to watch users of social networks in real time, according to contract documents.
The list of desired capabilities is impressive. The software is expected to do the following:
- “sentiment analysis,”
- “influencer identification,”
- “access historical Twitter data,”
- “detect sarcasm,”
- create “heat maps” or graphics showing user trends by color intensity,
- have “functionality to send notifications to users,”
- “synthesize large sets of social media data,”
- “identify statistical pattern analysis.”
What a plethora of perfectly phenomenal functions! Some tool it must be! With a piece of software like this Secret Service will be able to boost its effectiveness tenfold. After all, the agency is tasked to investigate any threats made against the president.
One of the requirements listed above makes this work order a tall order, however. It’s a requirement to develop an “ability to detect sarcasm.”
Even some reasonably intelligent human-folks have trouble detecting irony and understand humor sometimes. Sarcasm is especially tough to get.
It’s much easier in speech — uplifted brow, an eye-roll, a wink, a curve of a lip, or the speaker’s reputation of a sarcastic asshole give the listener subtle cues. Such cues cannot be embedded in text. Jokes often don’t translate well over SMS or Twitter.
Still, social media is full of it. I mean sarcasm. Every minute of every day, gazillions of sarcastic individuals exercise their right to carry the whole arsenal of undecipherable sarcasm. And they shoot it into the ether in either short bursts of tweets or in long barrages of blogs, comments and such.
“Historical Twitter data” Secret Service is eager to mine to “synthesize large sets of social media data,” is a landmine field with no warning signs “Attention! Sarcasm ahead!”
Computer program Secret Service is hoping for, must recognize genuine “threats made against the president,” alerting Secret Service to that effect but, at the same (real) time, totally disregard sarcastic verbiage and shit like that.
Is it even feasible?
To be sure, computer scientists have been diligently working on the problem. Oh yes, have they ever! It’s just that they haven’t yet been all that successful. Software happen to be stubbornly resistant to recognize sarcasm.
At their best performance, the computer programs used thus far could only correctly separate sarcastic from non-sarcastic tweets about 65 percent of the time — and this was in a highly controlled setting — the setting that is far removed from “real life” of a twittering blue bird.
I am not aware that anyone has a satisfactory algorithm or system that can detect sarcastic sentences,” said Bing Liu, a University of Illinois at Chicago computer scientist specializing in the field of sentiment analysis (that is, extracting emotional context from text).
Bing Liu expressed skepticism that anyone as yet has a good handle on this problem. To develop a software the Secret Service is seeking would be a particularly uphill battle:
“In discussions about politics [sarcasm] is fairly common and very hard to deal with because it often requires some background knowledge which computers are not good at.”
Jesse Singal in his article went all the way sarcastic:
So good luck to the Secret Service. Not just because sarcasm-detecting software could help keep the president safe, but because once it trickles down to the masses, it will save us all a lot of awkwardness in interpreting weird text messages. How helpful would it be if a little Bender popped up in the corner when your smartphone suspected your friend was just being a sarcastic jerk?