Ever wondered what scientists do on their off days? They create computer algorithms that can tell if you’re drunk while posting stuff on Twitter or Facebook. Apparently, the machine is so far off, that it can predict with an 80 percent accuracy you’re location and how drunk you are based on your text.
Leaving the joke aside, a team of scientists from the University of Rochester managed to shift the boundary of understanding deep-learning machines. Recently, the said team developed a highly sophisticated computerized algorithm that can predict with an uncanny accuracy your level of inebriation based on your social media posts.
If you’re wondering how those boys found out about the extra keg of beer you’re keeping under the couch in case your ex shows up, well things are quite tricky.
According to their statement, it seems that the first step they took in order translate a drunk tweet in machine language was to find some drunk Tweeters. Using geotags, the team tracked down several Tweets to New York City and Monroe County.
So, how did they figure the difference between a Tweet post written by someone who was sober, and a Tweet written by someone who was totally plastered? Quite simple. They’ve looked for words attributed to alcohol-related experience such as “party”, “keg”, “alcohol” or “drunk”.
After the first phase was over, the research team asked the help of Amazon’s Mechanical Turk in order to analyze more than 10.000 alcohol-related Tweets.
What were the results of their endeavor?
The in-depth analysis helped the computer to differentiate between a person posting about alcohol and a person who was drunk while tweeting. Thus, the team was able to find out that most of the so-called drunk tweets were coming from homes rather than pubs.
In the end, the team subjected their algorithm to a little test. Given a series of tweets, the computer had to find out if the person tweeting was drunk, and where they were posting from.
To their surprise, they’ve discovered that most drunken tweets originated from New York City, and most of the people sending messaged after having too much to drink were actually tweeting from home.
However, the scientists also added that the algorithm does not prove that New Yorkers drink more than other than the inhabitants of other cities. The goal of the project was to study how the inebriation process looks like from an online perspective and how social media pressure can affect one’s drinking habits.
So, next time you’re drunk, and you decide to post something utterly useless on Twitter, remember there might be someone at the end of the line laughing out loud.