Forecasting Virus Outbreaks With Social Media Data Via Neural Ordinary Differential Equations

This article examines the possibility of using real-time social media data as an early predictor of a new epidemic wave during the Covid-19 pandemic. A neural ordinary differential equation (neural ODE) was trained to forecast viral outbreaks in a specific geographic region using multivariate time series of signals derived from a novel set of large online polls regarding COVID-19 symptoms. The neural ODE was able to capture the dynamics of interconnected local signals and effectively estimate the number of new infections up to two months in advance. This study provides evidence for the predictive ability of widely disseminated social media surveys for public health applications.