People say some pretty ridiculous things when they're drunk. Years
ago, people only really had to worry about making an ass of themselves
around friends and family but now with outlets like Twitter, virtually
anyone can read your liquid courage-influenced ramblings.
It's this type of behavior that inspired researchers at the University of Rochester in New York do a study on the matter. Specifically, they set out to create an algorithm that could detect tweets that were likely published by someone under the influence of alcohol and even try to determine if they were at home or out while drinking.
Between July 2013 and July 2014, Nabil Hossain and a few other fellow researchers collected geo-tagged tweets in select areas of New York state. They filtered tweets based on if they included a mention of alcohol (those with keywords like "drunk," "beer," party" and so on).
The team then hired Amazon Mechanical Turks to apply a series of three questions to each tweet. Data from those answers was then used to create three different algorithms. As it turns out, the machine came to the same conclusion about a particular tweet as the human turks did between 82 percent and 92 percent of the time.
As Ars Technica notes, the researchers took it a step further to determine if a tweet was sent from a user's home or elsewhere. They did this by first filtering keywords (like "sleep," "TV, "bath" and so on) then again asked turks to comb through the tweets and see if they thought they were sent from home or not. This data was used to create another algorithm which they claim can determine a user's home location at an accuracy rate of up to 80 percent.
Just think, the next time you send a drunk tweet, you may end up being part of a large scientific study without even realizing it.
It's this type of behavior that inspired researchers at the University of Rochester in New York do a study on the matter. Specifically, they set out to create an algorithm that could detect tweets that were likely published by someone under the influence of alcohol and even try to determine if they were at home or out while drinking.
Between July 2013 and July 2014, Nabil Hossain and a few other fellow researchers collected geo-tagged tweets in select areas of New York state. They filtered tweets based on if they included a mention of alcohol (those with keywords like "drunk," "beer," party" and so on).
The team then hired Amazon Mechanical Turks to apply a series of three questions to each tweet. Data from those answers was then used to create three different algorithms. As it turns out, the machine came to the same conclusion about a particular tweet as the human turks did between 82 percent and 92 percent of the time.
As Ars Technica notes, the researchers took it a step further to determine if a tweet was sent from a user's home or elsewhere. They did this by first filtering keywords (like "sleep," "TV, "bath" and so on) then again asked turks to comb through the tweets and see if they thought they were sent from home or not. This data was used to create another algorithm which they claim can determine a user's home location at an accuracy rate of up to 80 percent.
Just think, the next time you send a drunk tweet, you may end up being part of a large scientific study without even realizing it.
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.