As reality and social media become ever more intertwined, the potential for real life embarrassment to result from drunken online activity grows ever higher. If you thought quickly hiding drunk photos from your employers and deleting cringe-worthy maudlin ramblings before your exes can see them does the trick, there's some bad news: Scientists have developed algorithms that can detect drunk tweets and geo-locate the users who wrote them, and they used it to monitor the drunken behavior of New Yorkers in 2014.
In what seems like a beta version of technology that will be fit for some future prohibition era, researchers at the University of Rochester developed a machine-learning algorithm that detects tweets relating to booze and can identify whether the tweeter was drinking when it was written, according to MIT Technology Review. A second algorithm can identify the location of tweeters' homes "with great accuracy," allowing the researchers to know whether the person tweeting was at home or not when they tweeted.
The researchers examined 11,000 geo-located tweets sent from January to July 2014 in both New York City and Monroe County, which is home to the University of Rochester. The team searched for words like "party," "drunk," and other words related to getting wasted. The researchers then asked three human users of Amazon's Mechanical Turk service whether or not the tweet related to alcohol and whether it was sent at the time of drinking.
Then the research team examined the subset of tweets for words like "sofa," "TV," or "Finally home!," and asked another set of Amazon Mechanical Turk users whether the tweet was sent from home. Putting it all together, the researchers developed an algorithm that can identify and map drunk tweets, distinguishing between home boozing and boozing out. Lead researcher Nabil Hossain says that his team's algorithms can now detect the location of a person's home to within 100 meters with up to 80 percent accuracy.
The research team says that the data is intended to study drinking habits and how it is distributed throughout society, and that it could be used to better understand public health concerns. Though perhaps skewed towards younger drinkers, the algorithm is cheap compared to traditional surveys and studies about drinking habits, which have their own biases such as underreporting or a reluctance to fill out surveys, particularly among some immigrant population
"We can explore the social network of drinkers to find out how social interactions and peer pressure in social media influence the tendency to reference drinking," Hossain said.
For now, we are left with some insight into the drinking habits of New York residents from two rather different cities. Using the data generated, the researchers were able to create a drunken heat map of New York City and Rochester.
Perhaps unsurprisingly, New York City residents drink more than their Rochester counterparts, and the heat map shows a lot of drunk tweets in downtown Manhattan and Williamsburg and its surrounding neighborhoods. And it turns out that drinkers are social butterflies in Monroe County, while New Yorkers were a bunch of drunk hermits. New York City residents were more likely to drink at home.
But if you are drinking out and tweeting about it for some reason, maybe the fact that a computer can tell if you're wasted from your tweets can serve as a little reminder to put down your phone.