Analyzing images on social media for (not) using mask

Abstract

The adoption of non-pharmaceutical interventions and their surveillance is critical for detecting and stopping possible transmission routes of COVID-19. A study of the effects of these interventions in terms of adoption can help shape public health decisions. Social media analytics can offer crucial public health insights. In this thesis we examined mask use and mask fit by analyzing 2.04 million posts from the 6 United States cities, 0.19 million posts during a large scale public event, Black Lives Matter (BLM) and compared how Democrats and Republicans differ in their behaviour towards NPIs, in the light of 2020 presidential elections using 3.9 million posts. The study finds a significant drop in the group posting when the stay-at-home laws were applied and a significant increase in mask wearing for two of the three cities when the mask mandates were applied. Although a general positive trend towards mask-wearing and social distancing is observed, a high percentage of posts did not adhere to the guidelines. BLM related posts were found to capture the lack of seriousness to safety measures through a high percentage of group pictures and low mask fit scores. Posts from Democratic states were found to have significantly higher percentages of detected masks, and significantly lower percentages of group photos, showing higher adoption of mask-wearing and social distancing measures than Republican states. Thus, the methodology used provides a directional indication of how government policies can be indirectly monitored through social media.