Machine Learning Can Tell Which Tweets Are Sent While Drunk

Back in the day before social networking, terms like drunk-dialing existed in which a person gets drunk and then calls someone, like an ex-girlfriend/boyfriend/wife/husband. Later when texting came along, drunk-texting became a thing, but now thanks to social networks like Facebook and Twitter, posting status updates while drunk is the new “in” thing.

That being said, have you ever wondered how to tell if a tweet was posted while under the influence of alcohol? Sure, sometimes it’s a dead giveaway with typos and exaggerated punctuation, but sometimes it’s not, which is why Nabil Hossain and his friends at the University of Rochester decided to train a machine to identify tweets that were sent out while the user was drunk.

For starters the method was straightforward in which tweets were filtered for words like drunk, beer, party, alcohol, and so on. Then they used Amazon’s Mechanical Turk crowdfunding service to further analyze the tweets to find out whether a tweet containing those words could be construed as a tweet sent while drunk.

While this might seem like a fun experiment, Hossain and his team do have bigger plans for this system in the future, such as how alcohol consumption might vary with age, sex, ethnicity, and so on, but for now it looks like a fun way to identifier users who might have tweeted under the influence of alcohol.

Tyler Lee
Categories: General
Tags: AI (Artificial Intelligence), Twitter