Or, more specifically, stomach complaints: nEmesis monitors diners' Twitter accounts for certain words that might indicate a potential food poisoning issue. Tests showed its findings closely matched those of health inspectors.
While at the University of Rochester, Adam Sadilek used machine learning to rank 3.8 million New York-area tweets according to the occurrence of words like “food” and “stomach.” Crowdsourced workers then identified 6,000 tweets that most likely signaled that the person doing the tweeting had experienced food poisoning. The set of data formed the key knowledge needed for Sadilek’s system, cleverly named nEmesis, to determine both the tweeters with upset tummies and the New York City restaurants that may or may not have been responsible. Further evaluation showed that the health scores assigned to restaurants by nEmesis closely matched those from city food inspectors.
What’s the Big Idea?
Perhaps not surprisingly, Sadilek now works at Google, which has its own service that uses search data to predict flu outbreaks. Using social media to predict larger trends has become increasingly common, but Sadilek readily admits that nEmesis is open for abuse by Twitter users: “People will start tweeting that they threw up when they know they are near McDonald’s.” He plans to present his system at the Conference on Human Computation & Crowdsourcing in November.