Analysis of the Public Debate from March to November 2020
Aggregating digital content streams from Web sites and social media channels, the Corona Mood Barometer uses automatic story detection and emotion analysis techniques to identify the latest trends. A Web-based visual analytics dashboard helps to better understand what drives the public debate and how government responses to the COVID-19 pandemic are perceived across the various countries.
RACE FOR THE COVID-19 VACCINE
Leading vaccine candidates have progressed through clinical phases at record speed. Pfizer / BioNTech, Moderna and AstraZeneca recently announced promising results from their phase 3 trials. The results show efficacy rates of more than 90% for both mRNA vaccines, and about 70% for the adenovirus-based vaccine of AstraZeneca. The news has accelerated efforts to win the race for the first vaccine rollout. The following video shows the evolution of media coverage since early November, highlighting the significant global impact of the recent announcements on the public debate.
The analysis shows how this has shifted global media attention. Journalists now focus on the authorization process, availability forecasts and fair distribution strategies. Differences in the required refrigeration temperature (Moderna at -20°C vs. Pfizer/ BioNTech at -70°C), for example, triggered online discussions that the lack of suitable fridges might slow down the distribution.
Which stories accompany us as many countries come out of lockdown? To answer this question, the Corona Mood Barometer provides a story graph visualization that automatically clusters content streams into a set of emerging stories (= groups of related news articles and social media postings), reflecting major events and intervention strategies to control the pandemic.
The story graph illustrates perceptions of the enacted lockdown and social distancing measures (terms in Italic appear as labels). While there has been widespread acceptance in many European countries, as shown in our previous analysis of the coverage in Austria, the situation is more divided internationally. Protesters complained about a perceived overreaction by health authorities and a threat to their personal freedom, as outlined in the section below. The first announcements of lockdown measures also brought shortages of various product categories including toilet paper, hand sanitizers and protective masks. Eyewitness accounts of the outbreak and a shortage of ventilators and intensive care beds in Italy and Iran highlighted the dramatic consequences that an exponential spread of the coronavirus can have. The rapidly increasing number of infections strengthened public support of government interventions. While the measures in Europe succeeded to flatten the curve, other countries have been less successful in containing the outbreak – including most recently Brazil and the United States. Partial reopening plans and the protests in Minneapolis against racism and police brutality have fueled the debate on the timing of reopening and whether the protests might contribute to a second wave.
GLOBAL CORONA MOOD OVER TIME
Text mining methods to classify content according to Plutchik’s Wheel of Emotions have identified five emotions particularly characteristic during this prolonged period of confinement: Anticipation, Vigilance, Fear, Anger and Sadness. The time series data of the Corona Mood Barometer shows clear peaks in these emotions, aligned with the most striking events in May.
The trend chart plots the mood deviation from the average since the first cases were reported among the residents of Wuhan City in January (measured in hourly intervals with a 24-hour moving average to smooth the trend line). This specific form of computation highlights the most characteristic events in the month of May. The low fluctuations in Fear and Anticipation indicate that these emotions already had a strong presence in the first quarter of 2020:
- “Anticipation” is closely linked with the global race for effective COVID-19 treatment options such as Remdesivir and Hydroxychloroquine or potential vaccine candidates such as Moderna’s mRNA-1273.
- “Vigilance” shows phases of slow decline, interrupted by official interventions or events – in the case of the changed UK government slogan “Stay Alert” (from the previous “Stay at home”), for example, or during the mass protests in the United States towards the end of the month that triggered a strong increase across several emotional categories.
- “Fear” remains on a rather constant level, with notable increases in mid-May when discussions about a second wave intensified and White House officials were tested positive for coronavirus, and towards the end of May due to fears of increased transmission of the virus during the mass protests.
- “Anger” exhibits several pronounced peaks on account of disagreement with stay-at-home orders and the management of the pandemic by governments and official health organizations such as the CDC. In the UK, the breaking of lockdown rules by Dominic Cummings, chief adviser to prime minister Boris Johnson, incited public anger, while Hong Kong saw major rallies against Beijing’s new national security law.
- “Sadness” increased during the 75th anniversary of VE Day, remembering the fallen of World War II, which coincided with coverage about the US unemployment rate being at the highest level since the Great Depression. Further peaks relate to the grim milestones of 300,000 global COVID-related deaths (May 14) and 100,000 in the US (May 27), as well as a plane crash after reopening the Karachi airport with almost 100 casualties.
DRIVERS OF THE GLOBAL CORONA DEBATE
Additional visual tools can be used to further explore the associations with each of the listed emotions, helping to better understand what drives the public debate. The tag cloud sorts major associations alphabetically, color coding them by emotion. The radar chart projects the top keywords along the chart’s multiple axes, revealing the relative strength of association with each emotional category. The keyword graph applies a hierarchical layout, the grey nodes in the center represent keywords that are linked to multiple emotional categories.
INTEGRATED “TOPIC COMPASS” DASHBOARD
All of the above visual tools are brought together into “Topic Compass”, a Web-based visual dashboard powered by webLyzard’s big data platform. The underlying algorithms are jointly developed with MODUL Technology as part of the research projects ReTV (“Re-Inventing TV for the Interactive Age”) and EVOLVE (“Leading the Big Data Revolution”), funded by the European Union’s Horizon 2020 Programme.
In addition to the presented international results, the platform has previously been used to analyze the public debate in Austria – in collaboration with the communications consultancy Ketchum Publico in the research project EPOCH (funded by the Austrian Federal Ministry on Climate Action, Environment, Energy, Mobility, Innovation and Technology). Whereas the conversation in Austria has also focused on policies, the appropriateness of measures and police enforcement of these measures, a decreasing number of new infections and the gradual easing of confinement has brought a glimmer of hope and joy.
- Nixon, L., Fischl, D. and Scharl, A. (2019). “Real-Time Story Detection and Video Retrieval from Social Media Streams“, Video Verification in the Fake News Era. Eds. V. Mezaris et al. Basel: Springer. 17-52.
- Scharl, A., Hubmann-Haidvogel, A., Göbel, M., et al. (2019). “Multimodal Analytics Dashboard for Story Detection and Visualization“, Video Verification in the Fake News Era. Eds. V. Mezaris et al. Basel: Springer. 281-299.
- Scharl, A., Hubmann-Haidvogel, A., Jones, A., Fischl, D., Kamolov, R., Weichselbraun, A. and Rafelsberger, W. (2015). “Analyzing the Public Discourse on Works of Fiction – Automatic Emotion Detection in Online Media Coverage about HBO’s Game of Thrones”, Information Processing & Management, 52(1): 129-138.
- Weichselbraun, A., Gindl, S., Fischer, F., Vakulenko, S. and Scharl, A. (2017). “Aspect-Based Extraction and Analysis of Affective Knowledge from Social Media Streams“, IEEE Intelligent Systems, 32(3): 80-88.