Dissemination of Vaccine Misinformation on Twitter and Its Countermeasures

Outbreaks of vaccine preventable diseases have continued to affect many parts of the United States. Measles particularly made a striking comeback in 2019, resulting in the greatest number of cases ever seen since it was declared eliminated two decades ago. The majority of the cases were among the under-or un-vaccinated, and whose vaccination status was mostly due to parents’ or own personal beliefs. While the causes of the growing vaccine hesitancy are likely to be multifactorial, the prevalence of misinformation on social media arguably plays a crucial role. To combat vaccine misinformation in today’s digital world, policymakers and other stakeholders need a more complete picture of the production, dissemination, and consumption of misinformation on social media.

This dissertation employed state-of-the-art machine learning models to collect the first dataset of vaccine misinformation from Twitter that was disseminated from January 2018 to April 2019. Out of 1,721,528 vaccine-related tweets, it was estimated that 15% were not credible, 11% were not supported by evidence, and 18% were considered propaganda. Consistent with anti-vaccine literature, topics about vaccine safety dominated the misinformation landscape. Nonetheless, vaccine misinformation also emphasized vaccine “truth”, using pseudoscience, spun presentation of legal proceedings, and “whistleblower” testimonies to mislead the public and sow doubt in the medical and scientific establishment. Besides, disseminators of vaccine misinformation took advantage of controversial issues such as abortion to incite anti-vaccine sentiment and grow their follower base. Unsurprisingly, most of the top sources appeared to have vested interests in exploiting false information to advance their financial or status gains. Contrary to common perception, however, most vaccine misinformation was disseminated by dedicated human actors as opposed to social bots. Fifteen percent of those who disseminated misinformation were likely bots and four percent were possibly completely automated. Finally, I proposed plausible actions that can be taken by social media platforms, the government, domain experts, as well as public health allies including clinicians and web influencers, and discussed generalizability of the analytical framework implemented in this dissertation.