Scientists Have Created a Machine That’s Very Good at Detecting Your Online Sarcasm
Now if only other people could learn to do it, too.
Sarcasm may be the official dialect of the internet, but it’s also a tricky beast, leaving misunderstandings and hurt feelings in its wake. Can a machine do what humans can’t—detect sarcasm in written communications?
That’s the motivation behind new research from scientists at Carnegie Mellon University, which finds that algorithms devised through social media sites like Twitter can actually figure out if web users are being sarcastic.
David Bamman and Noah A. Smith created an algorithm that trains itself by looking for social media posts using the hashtag #sarcasm. Over time, the algorithm became better and better at finding online sarcasm. That became truer as the researchers allowed their models access to more contextual information about social media users, like the topics they usually post about, whether the user had interacted with the subject of their post before, and who the user usually talks to online.
In the end, using all social media information available, the algorithm became pretty excellent at detecting sarcasm, finding it 85 percent of the time.
These results make sense: Sarcasm makes sense only in context, if people have certain sorts of information about each other. My friends know I’m being sarcastic when I text, “Let’s totally go to a rave tonight,” because they know I much prefer sleeping. Now, it seems, the computer might know it, too.