Twitter Tracker Shows Where Food Poisoning Is a Risk
New Scientist (08/06/13) Hal Hodson
A new machine-learning system can track tweets and show which restaurants are giving people food poisoning. Adam Sadilek built the system, called nEmesis, when he was at the University of Rochester. From January to April, the machine-learning system was used to rank a pool of 3.8 million tweets sampled in New York for relevant words such as stomach and food, and then a crowd of workers from Amazon's Mechanical Turk labeled the 6,000 most promising tweets on whether the tweeter was likely to have food poisoning or not. The system looked for when a person tweeted from a city restaurant, and monitored the account for a few days for keywords such as "throw up," "my tummy hurts," or "pepto-bismol." The program then assigned a health score based on the tweets, which closely matched the official score from the city's food inspectors, and found 120 restaurants that seemed responsible for at least one bout of sickness during the trial. Sadilek, who is now at Google, says Twitter users might try to abuse the system now that they know it exists. For example, he says, "people will start tweeting that they threw up when they know they are near McDonalds."
View Full Article
