With a drunk text, you always run the risk of embarrassing yourself in the eyes of whoever is on the receiving end of the message. But with a drunk tweet, you run the risk of embarrassing yourself to an untold number of followers. Worse yet, if a drunk tweet of yours goes viral, well, you may be forever associated with an ill-conceived message resulting from a long night spent out on the town.
To help prevent drunk tweeting, or perhaps to more readily identify drunk tweets once they go live, researchers at the University of Rochester recently came up with a machine learning algorithm that can purportedly identify tweets which were sent out in a drunken stupor.
The algorithm was crafted by first harvesting over 11,000 geotagged tweets from two areas in New York. The captured tweets all contained references, in some form or another, to alcohol related terms such as “beer”, “keg”, and “party.” Following that, the team of researchers took advantage of Amazon’s Mechanical Turk crowdsourcing service to ascertain which tweets were merely talking about drinking and which tweets were likely to have been composed by an intoxicated user.
Not only that, but researchers also came up with a way to determine when drunk tweets were being sent out from home or from a location elsewhere. They were able to accomplish this by keeping an eye out for certain phrases that tend to be used at home and outside the home. For instance, a tweet that says “finally back” tends to indicate a night out on the town while a tweet that references “taking a bath” points to a night spent drinking at home.
You can read more about this enterprising research over at MIT Technology Review.