![Drunk tweeting](https://www.siliconrepublic.com/wp-content/uploads/2016/03/Drunk-tweeting-faces-718x523.jpg)
Drunk tweeting has become an activity many have engaged in, and many have seen on their timeline, but now a new AI algorithm is being used to learn when you’re doing it based on your location.
Drunk tweeting has become something familiar to many, when your walk or taxi home seems like the perfect time to whip out your phone and send a series of tweets that you’ll likely regret the next morning.
The phenomenon even made headlines late last year as singer Adele admitted that her own Twitter account, which has millions of followers, is now monitored to prevent her from sending any impromptu tweets under the influence.
Now, to make matters potentially worse for those of you who engage in drunk tweeting, a team of researchers has spent an awful lot of time developing an AI algorithm that is learning when someone is sending a drunk tweet based on their location.
![Heat map drunk tweets](http://img.techxplore.com/newman/csz/news/800/2016/56ec151ca02f7.png)
Heat maps of drunk tweets using the algorithm. Image via arXiv:1603.03181
Could be used for police checkpoints
According to Tech Xplore, the researchers from the University of Rochester in the US have published a scientific paper on their algorithm, which was developed by collecting thousands of geotagged tweets from New York City and sending them to Mechanical Turk, an Amazon-owned on-demand service for businesses.
Once it was sent to Mechanical Turk, the researchers received information from people who had sifted through the tweets and guessed whether they were drunk tweets or not.
Once this data was fed into their algorithm, the researchers began tweaking it to be able to determine where the person might have been when they were drinking, based on the use of keywords like ‘bar’ and ‘nightclub’.
Once the researchers got everything they needed, they claimed the algorithm was able to locate where a drunk tweet was sent with an estimated accuracy of 80pc within 1km.
The research team has suggested that this type of AI development could be used to help police determine where to set up testing stations for drivers under the influence.