New Scientist (07/26/12) Michael Reilly
University of Rochester’s Adam Sadilek and colleagues were able to predict when individuals in New York City were about to come down with the flu up to eight days before they showed symptoms, using artificial intelligence and Twitter data. The team analyzed 4.4 million tweets tagged with global positioning system location data from more than 630,000 users in the New York City area over one month in 2010. The researchers trained a machine-learning algorithm to distinguish between tweets such as “I’m so sick of this traffic!” and those by people who were actually sick and showing signs of the flu. They were able to predict when someone was about to fall ill–and then tweet about it–with about 90 percent accuracy up to eight days in the future. Still, Sadilek says the system is limited because people do not reliably tweet about their symptoms and because getting sick is not limited to who they come in contact with. Nonetheless, the data from the system could potentially be used for a smartphone app that warns users when they are entering a public place with a high incidence of flu.