A team of international researchers, led by experts from Harvard University, has developed an algorithm harnessing diverse data sources—including social media—to forecast COVID-19 trends and inform policy decisions.
While COVID-19 shares similarities with past pandemics, its rapid spread caught governments worldwide off guard. In France and beyond, initial responses were delayed and inconsistent, fueling fears of a second wave. As authorities prepare for future actions, timely data-driven insights are crucial.
In a July 3, 2020, preprint on arXiv, the team detailed their forecasting algorithm, designed to predict pandemic shifts with a two-week advance notice. This window could enable proactive measures like deconfinement, recontainment, school reopenings, or economic restarts.
Lead researcher Mauricio Santillana from Harvard explained that the model detects risks 14 days before case surges by analyzing real-time Twitter activity, Google search trends, and smartphone mobility data.
The approach focuses on early behavioral shifts among populations. As the researchers noted, “early signals of increasing COVID-19 prevalence” emerge from these “next-generation alternative data sources.” For instance, Twitter mentions spiked a week before New York’s case explosion.
This research, published on arXiv before peer review, warrants cautious optimism. The authors acknowledge key limitations, such as the inability to predict abrupt societal changes—evident in events like the George Floyd protests and Black Lives Matter movement.
Over time, social media and searches may lose predictive power as populations adapt to the virus, reducing relevant queries. The team advises against current deployment due to unproven reliability, emphasizing the need for further validation.
