Miserable tweets, virus parameters help researchers accurately track the flu

twitter flu tracking
Greg Grinnell

Flu season is just about over in the United States, and while some people were suffering, researchers at Northeastern University were using social media to track the virus in real time.

By analyzing tweets with a computational model and combining that data with parameters about the virus, they were able to follow how the flu spread and forecast its development up to six weeks ahead of time.

“As modern transportation facilities have boosted human mobilities, the risk of infectious disease epidemics or pandemics is increasing,” Qian Zhang, the first author of the study, told Digital Trends. “In particular, seasonal influenza results in millions of illness and thousands of deaths every year world widely … Accurate in-time reporting of seasonal flu would help public health agencies, as well as health care units to be better informed and prepared. Unfortunately, the current surveillance data lag behind flu activities.”

With their new platform, Zhang and his colleagues looked to develop models that could help public health agencies better understand and predict how epidemics evolve. By incorporating GPS locations from Twitter users who were broadcasting their symptoms, the researcher were able to determine the severity of the epidemic in locations around the country.

The Northeastern team created their model in response to the “Predict Influenza Season Challenge” raised by the Centers for Disease Control and Prevention in November 2013. Although their model stood out from others thanks to its ability to predict certain dynamics of the virus — such when and to what extent the epidemic would peak — Zhang said, “We do like to think of the challenge as collaborative work with all the other teams.”

This research isn’t done yet though. Moving forward the team will continue to analyze epidemics, add new parameters, and turn to Twitter for details. “The evolution of epidemics can be quite different season by season,” Zhang said. “It means we have to explore a large parameter space and simulate thousands of models, from which we are able to identify an ensemble of possible models better describing flu epidemics for each season.”

A paper detailing the model was recognized at the 2017 International World Wide Web Conference in April.