Words Speak Louder Than Numbers: Estimating China’s COVID-19 Severity with Deep Learning

Abstract

We develop a deep learning algorithm to estimate the severity of the 2020 COVID-19 outbreak in China by analyzing the language of the People’s Daily, China’s official newspaper. The algorithm uses the 2002–2003 SARS outbreak as the benchmark and learns how the newspaper’s language evolved during the epidemic cycle. It then maps the daily coverage of the coronavirus outbreak to the SARS timeline and, hence, estimates its relative position in the benchmark epidemic cycle. We call this timeline-based measure the Policy Change Index for Outbreak. We find a pronounced discrepancy between our severity measure and China’s official numbers of diagnosed cases. We also demonstrate that our indicator is more informative about the outbreak’s severity than a conventional sentiment analysis.

Keywords: policy change, propaganda, deep learning, coronavirus, COVID-19, SARS, outbreak

JEL Classification: C63, D83, I18, P49