Improved healthcare disaster decision-making utilizing information extraction from complementary social media data during the COVID-19 pandemic.

Managing an extreme event like a healthcare disaster requires accurate information about the event's circumstances to comprehend the full consequences of acting. However, information quality is rarely optimal since it takes time to determine the information of relevance. The COVID-19 pandemic showed that even official data sources are far from optimal since they suffer from reporting delays that slow decision-making. To support decision-makers with timely information, we utilize data from online social networks to propose an adaptable information extraction solution to create indices helping to forecast COVID-19 case numbers and hospitalization rates. We show that combining heterogeneous data sources like Twitter and Reddit can leverage these sources' inherent complementarity and yield better predictions than those using a single data source alone. We further show that the predictions run ahead of the official COVID-19 incidences by up to 14 days. Additionally, we highlight the importance of model adjustments whenever new information becomes available or the underlying data changes by observing distinct changes in the presence of specific symptoms on Reddit.