Tracking the international anti-vaccination networks and their politics

Video: https://video.linux.it/w/xq3Z9khHJGTzv2NsNc1gK7?start=21m49&stop=37m29

I dibattiti anti-vaccinazione pervadono i social, alimentando la diffidenza verso la scienza e aumentando lo scetticismo riguardo ai vaccini.

Mentre gli studi precedenti si concentravano su paesi specifici, la pandemia COVID-19 ha ampliato il discorso sulla vaccinazione a livello mondiale, accrescendo la necessità di affrontare i flussi di informazioni poco credibili su scala globale per progettare contromisure efficaci.

In questo intervento descriverò come abbiamo utilizzato 316 milioni di tweet relativi alle vaccinazioni in 18 lingue per quantificare i flussi di disinformazione tra gli utenti esposti a contenuti anti-vaccinazione.

Scopriamo che, durante la pandemia, le comunità no-vax sono diventate più centrali nei dibattiti specifici per ogni paese e le loro connessioni transfrontaliere si sono rafforzate, rivelando una rete globale anti-vaccinazione.

Inoltre, abbiamo riscontrato una relazione complessa tra gli interessi politici degli utenti e la probabilità che condividano materiale anti-vaccini: gli utenti che seguono politici di partiti di destra e quelli associati a posizioni autoritarie o anti-UE hanno maggiori probabilità di essere esposti a contenuti anti-vaccini.

Yelena Mejova — Senior Research Scientist in ISI Foundation

Transcript

1. Tracking the international anti-vaccination networks and their politics Yelena Mejova | ISI Foundation Dec 13, 2023
2. https://www.cbc.ca/news/politics/cost-of-covid-19-misinformation-study-1.6726356 https://abcnews.go.com/Health/300000-us-covid-deaths-averted-vaccination-analysis-finds/story?id=847532
3. has COVID-19 changed our view of vaccination? (on Twitter)
4. Twitter vaccine debate in Italy •Twitter Streaming API •Italian language filter •Sep 5, 2019 – Nov 7, 2021 •>16M users, 665K tweets •6 time periods •6 retweet “endorsement” networks (weight > 1) Crupi et al. Echoes through Time: Evolution of the Italian COVID-19 Vaccination Debate. ICWSM 2022
5. Opinion communities •hierarchical clustering + selection using modularity •manual annotation of users •strong separation •pet users bridge hesitant and supporting camps hesitant pet supporting Crupi et al. Echoes through Time: Evolution of the Italian COVID-19 Vaccination Debate. ICWSM 2022
6. Almost nobody changes their mind hesitant pet supporting tweets per user per day Crupi et al. Echoes through Time: Evolution of the Italian COVID-19 Vaccination Debate. ICWSM 2022
7. … but they start mentioning each other more out of all mentions by row x, how many are from column y Crupi et al. Echoes through Time: Evolution of the Italian COVID-19 Vaccination Debate. ICWSM 2022
8. … and start talking about more similar things Crupi et al. Echoes through Time: Evolution of the Italian COVID-19 Vaccination Debate. ICWSM 2022 topical coverage from user group
9. What about the rest of the world? • Twitter Streaming API • Vaccine-related keywords in 20 European languages • 4 3-month periods • Twitter Academic API Historical failed to retrieve 72% of this data • Re-crawl users to find those suspended (A) 2020-8-11: Sputnik V vaccine announced (B) 2020-11-9: Pfizer-BioNTech vaccine announced (C) 2020-12-18: Moderna vaccine announced (D) 2021-1-4: First AstraZeneca vaccine inoculation. Lenti et al. Global misinformation spillovers in the online vaccination debate before and during COVID-19. JMIR Infodemiology 2023
10. Data • Geolocation using user location to GeoNames • Hand-check most popular matches and users • Keep countries with >2000 users in each period: 28 countries spanning 11 languages 20 Lenti et al. Global misinformation spillovers in the online vaccination debate before and during COVID-19. JMIR Infodemiology 2023
11. Community detection Lenti et al. Global misinformation spillovers in the online vaccination debate before and during COVID-19. JMIR Infodemiology 2023 0. Build a retweet network 1. Build the dendrogram of the hierarchical clustering using Paris algo 2. Compare the partitions obtained with cutorrs at heights 2, 3, 4, 5 3. Pick the partition with the highest modularity 4. If more than 90% of nodes are in the same community, compare the partitions with cutoffs at the following 5 heights and repeat from step 3 USA
12. Community labeling 0. Filter out communities having <1% of users 1. Sample 20 random tweets 2. Label tweets (in original language, and translated to English) as pro-vax, no-vax, or other 3. For communities having majority no-vax, label additional 10 most popular tweets 4. Community is “no-vax” if total number of no-vax tweets is >10 USA no-vax pro-vax Lenti et al. Global misinformation spillovers in the online vaccination debate before and during COVID-19. JMIR Infodemiology 2023
13. How do no-vax communities share info? • users in no-vax communities are more likely to retweet, share URLs, especially URLs to YouTube, and more likely to link to low-credible domains* * low-credible domains in Italian, English, French, and Greek Lenti et al. Global misinformation spillovers in the online vaccination debate before and during COVID-19. JMIR Infodemiology 2023
14. How is no-vax content moderated? • Users in no-vax communities are more likely to be suspended (except Cuba and Russia) • Lots of users are suspended immediately after the US Capitol riots Lenti et al. Global misinformation spillovers in the online vaccination debate before and during COVID-19. JMIR Infodemiology 2023
15. Cross-border spillovers: information • Normalized number of retweets (excluding diagonal) • English and Spanish language homophily (squares) • Germany/Netherlands & Germany/Turkey may imply expats • From pre-vax period, Russia is a net exporter, especially to South American countries Retweeting country Retweeted country Lenti et al. Global misinformation spillovers in the online vaccination debate before and during COVID-19. JMIR Infodemiology 2023
16. Cross-border spillovers: no-vax • Probability of interaction between users in no-vax communities from one country to another, w.r.t. interactions between other users from same countries • No-vax connections are generally stronger • English-countries, Germany, and Netherlands often together Retweeting country Retweeted country Lenti et al. Global misinformation spillovers in the online vaccination debate before and during COVID-19. JMIR Infodemiology 2023
17. Cross-border spillovers: low-credibility info • Proportion of URLs that come from retweeted country among the low-credible domains imported by retweeting country (countries importing less than 10 low-credible URLs are colored in grey) • US is global misinformation superspreader: 68% of all low-credible URLs retweeted worldwide come from US (a proportion much higher than the total volume (42%) retweeted from US) • Russia is central in exporting potential misinformation in the vax rollout period, especially to Latin American countries Retweeting country Retweeted country 27 Lenti et al. Global misinformation spillovers in the online vaccination debate before and during COVID-19. JMIR Infodemiology 2023
18. Cross-border spillovers: low-credibility info •Proportion of low-credible URLs imported from X, for each country and period considered. •Proportion of URLs that comes from U.S.: 93%, 79%, 74%, 56% •From Russia: 1%, 2%, 1%, 10% USA Russia 28 Lenti et al. Global misinformation spillovers in the online vaccination debate before and during COVID-19. JMIR Infodemiology 2023
19. What about politics? more likely to share no-vax • Following politicians from right-wing parties is usually associated with that account retweeting no-vax content • Following politicians from social democrat parties – less associated with no-vax • The relationship is stable over time Paoletti et al. Political Issue or Public Health: the Vaccination Debate on Twitter in Europe. ArXiv (under submission)
20. What about politics? more likely to share no-vax • Following politicians from right-wing parties is usually associated with that account retweeting no-vax content • Following politicians from social democrat parties – less associated with no-vax • The relationship is stable over time * over 17 European countries Paoletti et al. Political Issue or Public Health: the Vaccination Debate on Twitter in Europe. ArXiv (under submission)
21. but posts by politicians are no more influential than by other accounts with similar number of followers! What about politics? Paoletti et al. Political Issue or Public Health: the Vaccination Debate on Twitter in Europe. ArXiv (under submission)
22. Crupi, Giuseppe, Yelena Mejova, Michele Tizzani, Daniela Paolotti, and André Panisson. "Echoes through time: evolution of the Italian COVID-19 vaccination debate." International AAAI Conference on Web and Social Media (2022) Lenti, Jacopo, Yelena Mejova, Kyriaki Kalimeri, André Panisson, Daniela Paolotti, Michele Tizzani, and Michele Starnini. "Global Misinformation Spillovers in the Vaccination Debate Before and During the COVID-19 Pandemic: Multilingual Twitter Study." JMIR infodemiology 3 (2023) Paoletti, Giordano, Lorenzo Dall'Amico, Kyriaki Kalimeri, Jacopo Lenti, Yelena Mejova, Daniela Paolotti, Michele Starnini, and Michele Tizzani. "Political Issue or Public Health: the Vaccination Debate on Twitter in Europe." arXiv preprint (2023) yelena.mejova@gmail.com yelenamejova.com
23. We’re hiring! Getting your Masters degree by April 1, 2024? Work with us!