Exploring statistical relationships between social media sentiment and COVID-19 incidence rates

Abstract

In a crisis like the Corona pandemic, social media are often regarded as a mirror of society. With the help of the data obtained from such a social network, various methods can be used to determine how good the sentiment is on the respective network by deriving so-called sentiment scores from the messages of the users. This bachelor thesis aims to find out whether social media sentiments on Twitter can be explained by temporal patterns in COVID-19, such as case numbers, deaths and intensive care bed occupancy, taking into account general trends such as temperature and time itself. For this purpose various linear regression models were set up. First, regression models were set up for the federal state of Thuringia as a representative of a state with a fairly low amount of weekly tweets. Afterwards the same regression models were set up for the federal state of North Rhine-Westphalia, representing a state with a quite high amount of weekly tweets. The models are then analysed and compared to find out whether the differences in sentiment can be explained by the chosen factors and to ascertain whether there are differences between the two states.