What could possibly be more regrettable than drunk-texting a heart broken message to your ex after a night out? How about drunk tweeting said message for all to see?
Unfortunately for anyone who has ever become tweet happy after one-too-many, researchers at the University of Rochester have designed a computer algorithm that can recognize tweets that may have been sent under the influence.
The machine-learning program uses keywords associated with alcohol consumption, coupled with location data to predict whether a user is drunk tweeting.
Study author Nabil Hossain collected over 11,000 tweets between January and July 2014, geo-tagged within New York City and Monroe Country.
The software then sorted through tweets that contained keywords associated with drinking – like “beer,” “party,” “bar,” or “drunk.”
Researchers then used Amazon’s Mechanical Turk, a human crowdsourcing service, to further analyze the tweets. Each tweet was analyzed by three humans who were asked if the tweet made reference to drinking booze, if the tweet was about his or herself drinking and if the tweet was sent at a time and place where the user could have been drinking.
According to these human helpers, the algorithm was very accurate at detecting drunk tweets.
But Hossain decided to take it one step further – he set out to determine whether these alleged intoxicated tweeters were drinking at home, or out at a bar.
The algorithm filtered through the tweets to find words like “bath,” “sofa,” “TV” and “sleep” to determine if they were home when sending the tweets. Researchers were then able to map Twitter user’s homes on a map thanks to location data.
Sadly, it does not appear that there is a plan in place for this tool to be used to prevent people from drunk tweeting.
Researchers said they will use the algorithm to study how people consume alcohol and the social aspects of drinking – for example, how different settings influence people to drink and tweet.
“Our future work will perform a comprehensive study of alcohol consumption in social media around features such as user demographics, settings people go to drink-and-tweet,” the paper reads.
“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.”
© 2016 Shaw Media