The Role of Influencers and Echo Chambers in the Diffusion of Vaccine Misinformation: Opinion Mining in a Taiwanese Online Community.

Background: Prevalence and spread of misinformation are a concern for the exacerbation of vaccine hesitancy and a resulting reduction in vaccine intent. However, few studies have focused on how vaccine misinformation diffuses online, who is responsible for the diffusion, and the mechanisms by which that happens. In addition, researchers have rarely investigated this in non-Western contexts particularly vulnerable to misinformation. Objective: This study aims to identify COVID-19 vaccine misinformation, map its diffusion, and identify the effect of echo chamber users on misinformation diffusion on a Taiwanese online forum. Methods: The study uses data from a popular forum in Taiwan called PTT. A crawler scraped all threads on the most popular subforum from January 2021 until December 2022. Vaccine-related threads were identified through keyword searching (n=5818). Types of misinformation, including misleading, disinformation, conspiracy, propaganda, and fabricated content, were coded by 2 researchers. Polarity was proposed as a proxy for measuring an individual's level of involvement in the echo chamber, one of the mechanisms responsible for the viral misinformation on social media. Factors related to information diffusion, including misinformation type and polarity, were then assessed with negative binomial regression. Results: Of 5818 threads, 3830 (65.8%) were identified as true information, and 1601 (27.5%) contained misinformation, yielding 5431 boards for analysis. Misinformation content did not vary much from other contexts. Propaganda-related information was most likely to be reposted (relative risk: 2.07; P<.001) when comparing to true information. However, the more polarized a user was, the less likely his or her content was to be reposted (relative risk: 0.22; P<.001). By removing the nodes with a high level of indegree, outdegree, and betweenness centrality, we found that the core network and the entire network demonstrated a decreasing trend in average polarity score, which showed that influential users contributed to the polarization in misinformation consumption. Conclusions: Although the forum exhibits a resilience to echo chambering, active users and brokers contribute significantly to the polarization of the community, particularly through propaganda-style misinformation. This popularity of propaganda-style misinformation may be linked to the political nature of the forum, where public opinion follows "elite cues" on issues, as observed in the United States. The work in this study corroborates this finding and contributes a data point in a non-Western context. To manage the echo chambering of misinformation, more effort can be put into moderating these users to prevent polarization and the spread of misinformation to prevent growing vaccine hesitancy.