Forecasting influenza outbreaks before they strike could help officials take early action to reduce related deaths, which total 290,000 to 650,000 worldwide every year. In a recent study, researchers say they have accurately predicted outbreaks up to two weeks in advance—using only the content of social media conversations. The findings could theoretically be used to direct resources to areas that will need them most.
A team at the Pacific Northwest National Laboratory in Washington State gathered linguistic cues from Twitter conversations about seemingly non-flu-related topics such as the weather or coffee. Based on this information, the researchers nailed down when and where the next flu outbreaks were likely to occur.
The investigators used a “deep learning” computer model that mimics the layers of neurons and memory capabilities of the human brain. Their algorithm analyzed how Twitter language style, opinions and communication behaviors changed in a given period and how such changes related to later reports of flu outbreaks.
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
“The beauty of the deep-learning model we use is that it considers emotions and linguistic clues over time to predict the future,” says computer scientist Svitlana Volkova, who led the study, which was published last December in PLOS ONE. Previous efforts to forecast flu outbreaks via the Internet—including studies that used Twitter and Wikipedia records and a project called Google Flu Trends—have scanned specifically for flu-related words. In contrast, Volkova's work examined 171 million general tweets and outperformed other models that were based exclusively on word searches or clinical data suggesting an imminent outbreak.
“Estimating flu in specific, localized populations pushes the limits of what we thought we could do [with social media], and it opens the door to new possibilities,” says Mark Dredze, a computer scientist at Johns Hopkins University, who was not involved in the new study.
Epidemiologist Matthew Biggerstaff of the U.S. Centers for Disease Control and Prevention cautions that we are still in “early days” when it comes to flu forecasting. But researchers are increasingly looking to the Internet to supplement official data, which are limited to a small proportion of actual cases because many infected individuals do not seek medical care. Furthermore, such a tool might one day help identify flu trends in regions where public health data are not available at all.
