Version of record (VoR) available under DOI: https://doi.org/10.1145/3472720.3483619; information about research data is available in the HuggingFace repository: https://huggingface.co/datasets/webimmunization/COVID-19-vaccine-attitude-tweets

Although Twitter is regarded as one of the most potent sources of vaccine-related disinformation, relatively little is known about how Twitter constructs timelines for individual users. In this work, we examine the composition of the Twitter timeline conditioned on the initial selection of friends. We illustrate our method by analyzing how the initial selection of friends impacts the number of pro- and anti-vaccination information present in the timeline. Our experiment clearly shows the disproportionate prevalence of anti-vaccination content seeping into Twitter timelines even for accounts initialized with explicitly pro-vaccination friends. We also discuss ethical considerations of using automated bots for research purses in a social network environment.