If you were at a bar with your phone in New York City during 2014, your whiskey-laced illegible tweets may have been used for important scientific research.
In a study out this week, a group of computer scientists from the University of Rochester analyzed over 11,000 geotagged drunk tweets posted in New York City and Monroe County in Upstate New York from January to July of 2014. How did they isolate the intoxicated ones from ones that were sober and just kinda stupid? The first part was easy, narrowing tweets for words like “beer,” “party,” “drunk” (and, I would hope, misspelled variations of these words as well). The team then enlisted Amazon’s Mechanical Turk for more qualitative analysis to determine if the tweet was sent at the same time the tweeter was drinking. Which for some of us would be anytime after 10:00 pm.
From this data the team developed a machine-learning algorithm to automatically find tweets that are composed while drunk. But it turns out that pinpointing where we decide to tell the internet how wasted we are can tell scientists much more than that. Using the drunk tweeting data it was relatively easy to find the location of a user’s home, for example, due to phrases that referenced things like “couch” and “finally back” and probably a quest for pizza. “Our results demonstrate that tweets can provide powerful and fine-grained cues of activities going on in cities,” said study author Nabil Hossain.
But the whole study wasn’t just for a giggle at the expense of sloshed social media users. Mapping the boozy tweets can show where people are drinking—or, more specifically, drinking too much—and could help cities point out important trends when it comes to alcohol policy and public health. As MIT Technology Review points out, this kind of research could prevent some of the 75,000 alcohol-related deaths in the US every year.
The serious nature of the study is probably also why the researchers didn’t include any of the sample tweets in their study, which is why the Gizmodo staff has taken it upon ourselves to show you, hypothetically, what these kinds of tweets might look like.