Even machines now know when you’ve been live-tweeting during The Walking Dead after half a bottle of wine.
Researchers from the University of Rochester developed a machine-learning algorithm that is able to spot drunken tweets. Researchers were able to do this by first collecting geo-located tweets from New York City and Monroe County in the state of New York and then filtering out tweets that contained alcohol-related words like “drunk,” “beer,” and “party.”
Researchers then teamed up with Amazon’s Mechanical Turk crowdsourcing service to analyze some 11,000 alcohol-related tweets. The researchers were able to distinguish whether the person was just tweeting about alcohol, or was actually drinking while tweeting.
Researchers were then able to pinpoint whether these people were tweeting from home, or within 100 meters of home, with 80% accuracy.
They found—surprise, surprise—not only where there more alcohol-related tweets in New York City than Monroe County, there were also more people drinking at home. That said, researchers acknowledge the limitations of their machine. Twitter is not a representative sample of the general population, and neither is New York City, so there is a clear bias in the data.
Researchers suggest that the study is an important tool to better understanding research into alcohol consumption, the impact of social media and peer pressure, and how alcohol consumption varies between different social groups.