Source-linked AI summary
COVID-19 Vaccine Hesitancy on Social Media: Building a Public Twitter Dataset of Anti-vaccine Content, Vaccine Misinformation and Conspiracies
Goran Muric, Yusong Wu, Emilio Ferrara
TL;DR
COVID-19 vaccine misinformation and anti-vaccine narratives threaten public health, while researchers lack accessible data focused on such content and accounts. The paper releases two complementary Twitter collections and descriptive analyses, totaling over 137 million tweets by early May 2021. The dataset provides a public resource for studying anti-vaccine misinformation and vaccine hesitancy through social media.
Problem
Accessible data focused specifically on anti-vaccine content, misinformation, conspiracies, and susceptible accounts’ historical Twitter activity is limited.
Method
The paper builds a keyword-centered streaming collection and a historical account-level collection, then analyzes activity, topics, news sources, geography, and political leaning.
Results
Over 137 million tweets were collected in total, including 1.8 million streaming tweets and more than 135 million historical tweets from approximately 70K accounts.
Takeaways & Limitations
The publicly released dataset provides research assets for studying anti-vaccine misinformation and vaccine hesitancy through social media.
Takeaways & Limitations
The keyword-based collection may miss evolving anti-vaccine topics and nuances, and Twitter users are younger and more politically engaged than the general public.
Abstract
from arXiv · showhide
False claims about COVID-19 vaccines can undermine public trust in ongoing vaccination campaigns, thus posing a threat to global public health. Misinformation originating from various sources has been spreading online since the beginning of the COVID-19 pandemic. In this paper, we present a dataset of Twitter posts that exhibit a strong anti-vaccine stance. The dataset consists of two parts: a) a streaming keyword-centered data collection with more than 1.8 million tweets, and b) a historical account-level collection with more than 135 million tweets. The former leverages the Twitter streaming API to follow a set of specific vaccine-related keywords starting from mid-October 2020. The latter consists of all historical tweets of 70K accounts that were engaged in the active spreading of anti-vaccine narratives. We present descriptive analyses showing the volume of activity over time, geographical distributions, topics, news sources, and inferred account political leaning. This dataset can be used in studying anti-vaccine misinformation on social media and enable a better understanding of vaccine hesitancy. In compliance with Twitter's Terms of Service, our anonymized dataset is publicly available at: https://github.com/gmuric/avax-tweets-dataset
Introduction
Vaccine hesitancy and anti-vaccine narratives pose public-health concerns, while social media facilitates misinformation and community formation. The paper addresses limited data access by releasing a large Twitter dataset focused on anti-vaccine content and related conspiracies.
- Vaccine hesitancy includes delaying or refusing vaccination despite available services, alongside safety, religious, access, unwillingness, and fear-related factors.
- Social media makes misinformation dissemination easier and enables anti-vaccine communities to form around shared sentiment.
- 1.8 million tweets come from a keyword-centered streaming collection, while more than 135 million historical tweets come from approximately 70K anti-vaccine accounts.
- The dataset is publicly released as anonymized tweet IDs in compliance with Twitter’s Terms of Service.
Methods
The paper constructs complementary Twitter collections using anti-vaccine keywords and historically active anti-vaccine accounts, then derives measures of hesitancy, political leaning, media sources, geography, and hashtag topics.
- Streaming collection: A snowballing procedure expands manually curated anti-vaccine seed keywords through co-occurring terms collected with the Twitter Streaming API.
- Account collection: Approximately 70K accounts identified in the streaming collection are sampled for historical tweet retrieval through the Twitter Search API.
- Avax score: The avax score measures the proportion of an account’s tweets containing anti-vaccine keywords among all its tweets.
- Political leaning: Political leaning is inferred from the political bias of media outlets and accounts shared through retweets and original tweets.
- Media sources: URLs are parsed into components, while known misinformation domains and Media Bias/Fact Check support media-source identification after excluding shorteners and major platforms.
- Geography: State activity is estimated from self-reported account locations and normalized by 2010 Census population, with no account-collection geolocation map because few accounts reported locations.
- Hashtag topics: The hashtag topic network links hashtags appearing in the same tweet, weights edges by co-occurrence, sizes nodes by degree centrality, and detects communities with Louvain clustering.
Data overview
The dataset combines a smaller real-time keyword stream with a much larger historical account collection. The overview reports their scale and emphasizes that collection remained ongoing.
- Over 137 million tweets had been collected in total by early May 2021.
- The streaming collection contains 1.8 million tweets, whereas the account collection is significantly larger because it covers historical activity from susceptible anti-vaccine accounts.
- The reported statistics may vary in future dataset versions because data collection was still ongoing.
- Table 2 presents basic statistics for the streaming and account collections.
Streaming collection
The streaming collection tracks anti-vaccine Twitter activity over time, geography, hashtags, topic communities, and shared news sources. Activity increased over the collection period, was concentrated in English-speaking countries, and contained recurring anti-vaccine, conspiracy, and misleading-information themes.
- Activity over time: 1.8 million tweets from 719K unique accounts were collected between October 18, 2020 and April 21, 2021, with activity generally increasing over time.Small spikes usually occurred around major vaccine research or authorization announcements, while a large late-November spike reflected increased activity from a small number of accounts.
- Geographical distribution: Approximately 68% of tweets originated in the United States, followed by Great Britain at 12.5% and Canada at 5.5%.Within the United States, California, Texas, Florida, and New York led in absolute volume, while Hawaii, Alaska, and Maine ranked highest after population normalization.
- Hashtags: The hashtag data included strong anti-vaccine terms, with #novaccineforme appearing in more than 25K tweets and representing 6.6% of hashtag-containing tweets.Other frequent hashtags referenced debunked depopulation conspiracies, while seemingly benign tags such as #learntherisk and #informedconsent often functioned as decoys.
- Topic communities: Three hashtag communities centered respectively on depopulation conspiracies, vaccine safety concerns, and a mixture of anti-vaccine, neutral, and pro-vaccine terms.The communities were identified using Louvain detection on a topic co-occurrence network.
- News sources: The top shared URLs were dominated by low-credibility websites, while NCBI links could lend scientific legitimacy to tweets presenting misleading interpretations of research.The paper describes cherry-picking rare adverse effects and overstating them as strategies used to support vaccine boycotts.
Account collection
The account collection provides a longitudinal view of anti-vaccine Twitter activity, revealing substantial historical coverage, right-skewed inferred political leanings, and frequent far-right or conspiratorial media sources.
- Over 135 million tweets from more than 78K accounts span March 3, 2007 to February 8, 2021.Approximately half of accounts published fewer than 1,500 tweets, while 40% published more than 2,000.
- Approximately 70% of accounts have their oldest collected tweet from 2020, while 18% date before 2018 and 6.8% before 2014.
- The most common hashtags include COVID-related and US-political terms, reflecting heightened activity around the 2020 presidential election.
- Political-leaning estimates inferred from both original tweets and retweets skew strongly to the right.The estimates are based on accounts’ media diets.
- The Gateway Pundit appears more than 180,000 times, while Breitbart News and the Epoch Times also occur frequently among account-collection URLs.Periscope tops the listed high-credibility URLs and was used to share political conspiracies and anti-vaccine narratives.
Discussion
The paper contributes two complementary Twitter collections for examining anti-vaccination narratives and characterizes them across content, sources, geography, and political leaning. It also identifies coverage and representativeness limits that constrain interpretation.
- The dataset combines a real-time keyword-based streaming collection with a longitudinal account collection of historical tweets.
- The authors characterize prominent keywords, news sources, geographical locations, and inferred political leanings.
- The streaming collection may miss evolving anti-vaccine terminology and nuanced forms of vaccine hesitancy because it relies on defined keywords.
- The dataset should not support conclusions about the general population because Twitter users are younger and more politically engaged than the public.
- The account collection supports further study of accounts engaging in anti-vaccine narratives and their longitudinal characteristics.
Usage Notes
The dataset is released under Twitter’s terms and policies, and its use requires researchers to follow the associated license and platform regulations.
- Researchers using the dataset must agree to its associated license and conform to Twitter’s policies and regulations.