Using machine learning, researchers at the University of Rochester have created a system that can find alcohol-related tweets and determine whether they were made by someone who was actually drinking at the time. It can also pick out whether those tweeters were drinking at home or somewhere else.
Equipped with that knowledge, the researchers compared the results for different locations in New York State. Eventually, they hope to use the technology to study the health implications of alcohol.
So, how did they do it
The researchers began by collecting geotagged tweets sent over 12 months up to July 2014 from New York City and nearby suburban and rural areas. Then they zeroed in on those that included alcohol-related words, such as "drunk" or "beer," and put the Mechanical Turk crowds to work to help confirm the context.
Specifically, Mechanical Turk workers read the tweets to confirm not only that the tweeter was talking about using alcohol personally, but also that he or she was consuming it while tweeting.
To figure out the settings from which the tweeters sent their tweets, the researchers focused on words and phrases that people tend to use in tweets sent from their homes, such as "bath" or "sofa," and confirmed the geolocated results once again via Mechanical Turk.
They used all that information to train a machine-learning algorithm, which they hope to use to better understand alcohol consumption patterns and how they vary with location and other factors. That data could help them do things like relate the number of places to buy alcohol in a region to the amount of home drinking that takes place there.
A paper describing the work will be presented at the International Conference on Web and Social Media in May.