Perhaps, in the future, breathalyzers will be replaced by the timeline of your Twitter account: a group of researchers from the University of Rochester, in the United States, led by Nabil Hossain, has designed an algorithm to try to find out if a tweeter you are sending your messages under the influence of drink .
Thanks to it, these experts say, useful statistical conclusions can be drawn to design more effective campaigns against alcoholism . And how have Hossain and his collaborators trained their “machine” to detect drunk tweeters?
First, they have analyzed 11,000 geolocated tweets in New York City and Monroe County that included words related to drinking: “drunk,” “beer,” “party,” and so on. To do this, they had the collaboration of so-called turkers , people who do tasks such as labeling data, texts or photos, who were asked to rate the messages and award them to “tipsy” users.
Then they also did that painstaking job of sorting to establish whether the tweeters were at home or using the social network from a bar or other place . The third step was to locate the tweets on a map and compare it with the existing statistics of alcohol consumption. This triple filter refined the algorithm and allowed several conclusions to be drawn, such as that tweeters in New York mostly drank at home and that Monroe users went out more often to do so , up to more than a kilometer away.
There was also a geographic correlation between the density of establishments that sell drinks and tweets related to alcohol. But in this case it is difficult to know if people consume more because there is more drink available or, vice versa , if people who are fond of alcohol choose to live in areas where there are plenty of stores that supply them with the merchandise.
The greatest usefulness of this type of algorithm is that it allows real-time monitoring of social habits and phenomena , when before it was necessary to design expensive and laborious methods, with selection of individuals, personalized surveys, etc.
In the future, researchers at the University of Rochester will try to further refine their algorithm to study alcohol use based on ethnicity, age, or gender, for example.
