Understanding the Anti-Mask Debate on Social Media Using Machine Learning Techniques

摘要

Masks are believed to slow the spread of Covid-19, and can prevent many deaths, yet this inexpensive, common sense public health measure has ignited a fierce debate in the United States. Opponents of masks or anti-maskers have resorted to measures such as organizing protests and marches to make their views public. They have also taken to social media platforms to vigorously argue against the use of masks. Even with the advent of vaccines, masks are still likely to be recommended for a long time. It then becomes important to mine the debate around masks to understand the concerns of the detractors and the arguments used by the proponents to counter these concerns. This paper analyzes the mask dialogue on Twitter, using the data collected in July and August 2020, which coincided with the time when the stay-athome orders were being relaxed, and the opening of schools and other activities was being contemplated. These tweets are explored in three ways – informal opinion mining is used to reveal the reasons for concerns and support, social parameters of the tweets and tweeters are analyzed to expose the dynamics of the two communities, and classification framework is built to distinguish between pro- and anti-mask tweets so that the latter can be tagged to prevent the spread of discordant information. Our results indicate that the concerns of antimaskers are more political and ideological rather than related to the adverse health impacts of masks. Members of the close-knit, small anti-mask community promote each other’s views compared to the pro-maskers, although the antimaskers themselves are not fringe by any means. The classification framework can detect anti-mask tweets with excellent accuracy of over 90%, and hence, it can be used to label tweets that sow misinformation about masks before they spread through the ether and influence people.