A team of researchers at the University of Rochester led by Nabil Hossain have trained a machine-learning system to recognize drunk tweets. It wasn't just for fun—they are using the data to research when people drink at home and when they drink other places.
The team explained that they first collected geotagged tweets from New York City and Monroe County, New York (where Rochester is located). They then used Amazon's Mechanical Turk service to hire people to identify which tweets were sent by people who were drinking at the time. That dataset allowed them to train the machine-learning algorithm to identify drunk tweets. They also hired people to identify which tweets users had sent from home. That allowed them to identify a home geotag for the tweeters, which they then used for research to determine when people were drinking at home and when they were drinking someplace else.
They plan to use their data and techniques for more research that could help prevent alcohol-related deaths.