Scientists can tell which tweets were sent by drunk users

sQuashie

March 18, 2016 11:42 AM

Researchers have developed an algorithm to map tweets under the influence of alcohol and establish recurring patterns in the behavior of Internet users.

Musician,

Nabil Hossain, a PhD student at the University of Rochester and an expert in machine learning, has developed an algorithm able to detect messages written under the influence of alcohol. The issue is not trivial because this model could provide new information on alcohol consumption.

According to MIT, Nabil Hossain has established a true mapping of drunken tweets of the New York.

His tool displayed alcohol outlets, home of the tweeters, and the location of the messages. In the end, his team was able to show that people were drinking mainly at home more in New York, and even closer to a place of sale of alcohol.

To achieve this, researchers conducted research in two stages. The first was to identify tweets that appear to have been written while intoxicated.

As a working method, they checked geotagged tweets from New York to Monroe County, through Rochester in upstate New York. From there, they selected messages which used words revolving around alcohol with the words "drunk", "beer", "party", etc ...

Then the application of crowdsourcing Amazon Mechanical Turk was used to recruit users to analyze tweets in detail and select only 11,000 messages that seemed really written under the influence of alcohol. Sufficient data to operate machine learning algorithms.

The team finally developed various methods to determine where and when such messages were sent.

Selection of words such as "home" or "sofa", marking the place where the first and last tweet of the day are sent, percentage of tweets sent from one place, and verification by employees of Mechanical Turk, were the many tricks that helped locate the home of Internet users with an accuracy of less than 100 meters.

Through Twitter and inexpensively, scientists have proved that we could better understand the consumption habits of the population.

They now plan to target new criteria such as age, sex or origin of users and even to try to determine the peer pressure in the drinking. Still only through the social network.

Enhance Your Pulse News Experience!

Get rewards worth up to $20 when selected to participate in our exclusive focus group. Your input will help us to make informed decisions that align with your needs and preferences.

I've got feedback!

sQuashie