Twitter helps to predict flu trends in New York

樵夫

Researchers from the University of Rochester in New York used Twitter to predict flu trends among individuals up to eight days before symptoms became present, according to a recent article in New Scientist. The researchers analyzed 4.4 million tweets over a one month time span in 2010 that were tagged with GPS location data from more than 630,000 users in New York City. After developing an algorithm that could differentiate between phrases such as "I am so sick of this traffic!" and actual declarations of illness, the researchers then sorted through the information and predicted with 90 percent accuracy whether an individual would become sick.

Similarly, earlier this year, researchers affiliated with Harvard Medical School and Massachusetts General Hospital used Twitter to forecast a cholera outbreak in Haiti two weeks before health officials in the country reported the epidemic

樵夫

If you've been walking around a public place lately, you've come in contact with a lot of people. Some of those people may have been sick. And if you've been hanging around enough of them as they cough and sneeze, then you might be about to get sick too.

That may sound obvious, but Adam Sadilek at the University of Rochester in New York and colleagues have applied the idea to a pile of Twitter data from people in New York City, and found that they can predict when an individual person will come down with the flu up to eight days before they show symptoms.

It's a similar idea to Google Flu Trends, which tracks how often people search for "flu" and related terms on the search engine and uses that information to provide daily updates on where outbreaks are occurring and how they're spreading.

To see whether it was possible to bring such a service down to the level of the individual, Sadilek and his team analysed 4.4 million tweets tagged with GPS location data from over 630,000 users in the New York City area over one month in 2010. They trained a machine-learning algorithm to tell the difference between tweets by healthy people - who might say something like "I am so sick of this traffic!" - and someone who is actually sick and showing signs of the flu. The video above shows a heat map of flu occurrence over the course of one day, based on their findings.

The researchers were able to predict when healthy people were about to fall ill - and then tweet about it - with about 90 per cent accuracy out to eight days in the future.

The system is limited in several ways. For instance, it misses many cases of illness because people don't reliably tweet about their symptoms. And there are many factors that go into people getting sick, not just who they've had contact with.

Sadilek is addressing some of these confounding factors. In unpublished findings described to New Scientist during an interview at the Conference on Artifical Intelligence in Toronto, Canada, yesterday, his team showed that people who go to the gym regularly are moderately less likely to get sick. People with low socio-economic status, on the other hand, are much more likely to become ill.

Such information could one day be used to power a smartphone app that warns you when you've entered a public place with a high incidence of flu. Or after a big day out, it might buzz you with a message to say you are at high risk of getting sick over the next few days.