Measuring and Validating Collective Emotions on a German Online Newspaper Platform in the course of the COVID-19 Pandemic (Pre-registration for Master Thesis)

With the proposed thesis we address two primary research questions in the context of text-based sentiment analysis on social media. First, we investigate whether the sentiment expressed in user postings on a German online newspaper platform positively correlates with sentiment variables collected from the representative COVID-19 Snapshot Monitoring (COSMO) study during the COVID-19 pandemic. With this correlation we aim to analyze whether expressed sentiment on social media reflects the collective emotions of society during a major crisis. Second, we will explore how the sentiment of user postings on a newspaper platform relates to the daily fluctuations in COVID-19 case numbers, providing insights into the pandemic's impact on public sentiment over time. To answer the research questions, we will employ three sentiment models, namely German Sentiment, a model for political text, and the DE-LIWC2015 dictionary. We will assess the correlation between the models' outputs and the COSMO data as well as COVID-19 case numbers. We hypothesize that user-posted sentiments correlate with COSMO data, reflecting society's collective emotions, and that negative sentiment patterns correspond with daily COVID-19 case fluctuations, albeit with an expected lag of one to three days. In addition, we explore the dynamics of collective emotions, examining how they respond to various pandemic waves and virus variants. We anticipate that the extent and rapidity of emotional shifts depend on the perceived threat of the virus variant, impacting the correlation with COVID-19 case numbers. We leverage data from Die ZEIT Online and COSMO, offering insights into the potential of social media sentiment to reflect and inform researchers and policymakers about societal emotions during crises. The COVID-19 pandemic provides a unique backdrop for our research, given the increased use of social media and the population's emotional state during the crisis. Our work aligns with ongoing sentiment analysis research, aiming to provide further insights. While challenges, such as sentiment analysis tool accuracy and alterations in user behavior, persist, the representative COSMO data provides an opportunity to validate our sentiment analysis results. Research questions: (1) Does the sentiment of user postings on a German online newspaper platform, as measured by three different sentiment models, positively correlate with variables assessing mood collected as part of the representative COSMO study, and therefore reflect the collective emotions of society during a major crisis, i.e., the COVID-19 pandemic? (2) How does the sentiment of user postings on a German online newspaper platform correlate with the daily fluctuations in COVID-19 case numbers, and what can this relationship reveal about the impact of the pandemic on public sentiment over time? Study methods: We will run an analysis using three different sentiment models, namely German Sentiment, a model for political text, as well as the dictionary DE-LIWC2015. To answer the research questions, we determine the correlation between the models' outputs and (1) the COSMO data and (2) the COVID-19 case numbers. Furthermore, we will calculate basic statistical parameters such as the average number of user postings per day, as spikes in activity may be related to important events such as the introduction of new restrictions and provide further insights into the behavior of the population online during the pandemic. Hypotheses: (1) The emotions expressed on social media by comparatively few individuals, in this case the actively posting users of Die ZEIT Online, positively correlate with sentiment data collected through a representative survey (COSMO), and therefore reflect the collective emotions of society to a moderate to high degree. (2) A 7-day rolling average of the collective emotions of Die ZEIT Online user postings correlates with an expected lag of a one to three days with the development of the 7-day rolling average of COVID-19 case numbers in Germany. (3) The intensity of the dynamics of the collective emotions, i.e. extent and rapidity of shifts from expressed positive to negative emotions, is dependent on the particular wave and the respective virus variant. This implies that the claimed correlation in (2) is less prominent if the prevailing virus variant is perceived less dangerous, compared to the case where the prevailing variant is considered more dangerous. Or in other words, the collective sentiment moves less in a negative direction as the number of cases increases.