Abstract:
The formation and maintenance of echo-chambers reinforced by social media’s ranking algorithms are the center of IS researchers’ interests. Nonetheless, there are few analytical tools that effectively analyze and visualize the topical changes, impeding our understanding of how topics emerge, morph, and be diffused within echo-chambers. The objective of this study is to introduce echo-chamber analytics and to apply it to identify the evolution of statements made in polarized groups on social media during a major societal event. As our testbed, we chose pro- and anti-vaccine groups on Twitter, as the long-standing online competition between the two groups has become fierce during the global COVID-19 pandemic. Especially, we focused on November - December of 2020 when pharmaceutical companies underwent the testing and approval of their immunizations and after the conclusion of a politically polarized US presidential that intensified the pro- and anti-vaccine echo-chambers. We then compared this November 2020 set with November 2019 to investigate topical changes, if any. We used a classification algorithm we previously developed to identify the pro and anti-vaccine echo-chambers, and k-clustering to group similar tweets together from each echo-chamber. 
 
Our dataset of November of 2019 and 2020 contained 11,103 vs 11,573 pro-vaccine and 8,168 vs 10,831 anti-vaccine tweets, showing a 32% increase in anti-vaccine tweets from 2019 to 2020 compared to only 4% increase in pro-vaccine tweets. While 4 clusters of each pro- and anti-vaccine clusters were found in 2019, only 2 clusters of each side emerged in 2020. Our inductive coding results show that pro-vaccine topics include (1) safety/effectiveness of the vaccine and (2) the protection/benefits of herd immunity , while anti-vaccine counterparts encompass (1) distrust in pharmaceutical companies and (2) safety concerns. We make both methodological and theoretical contributions: (1) echo-chamber analytics will facilitate the advancement of various IS theories on polarization and (2) the results obtained from echo-chamber analytics expands our understanding of changes in statements made on social media during a major social event involving two competing groups online.