Title: PCI-Outbreak: How Serious Has the Outbreak of COVID-19 Been in China? 

Abstract:

What do we know about the outbreak of COVID-19 in China? It has been suspected that there is an inconsistency between the reality and the official statistics. A model that reflects the real situation would be helpful to fight against the global pandemic here in the United States. To mine the truth about the pandemic with the limited available information, we developed an open-source, reproducible, and extensible model, called PCI-Outbreak, to gauge the severity of the outbreak in China by “reading” People’s Daily, the China’s state newspaper, with Policy Change Index (PCI). PCI has already proved to be successful in modeling other trends in China. It is powered by state-of-the-art neural-network-based and transformer-based language models.

We trained our model using the news articles in People’s Daily during the SARS outbreak in 2003. We assume that the Chinese government officials react in a similar manner when dealing with COVID-19, resulting in a similar language style on the state newspaper. With PCI-Outbreak, we found that the severity of the pandemic has dropped at a much slower rate than it has been officially announced. Despite the optimistic tone from the government, the real situation may be much more serious. While the official statistics do not reflect the reality, the machine can still deduce the severity of the outbreak from the language styles. Our model is also extensible to other potential outbreaks, helping our policy makers take early precautions in the future.