Counteracting disinformation
5 min

Anti-vaxxers. Bots vs Opinion makers

showcase image
Counteracting disinformation

About the project

Note: This project was conducted as part of Artellence and, together with similar projects, became a prerequisite for the creation of the NGO Association for the Development of Big Data and Information Technology of Ukraine with the involvement of experts who directly implemented this project.

62% of Ukrainians get their news from social media. In the midst of deciding whether or not to get vaccinated against COVID-19, Ukrainians' decisions are influenced by information from social networks.

The purpose of the study is to analyze how many posts and comments on Facebook are written by people who do not sincerely believe in science or by managers of bot farms who are fulfilling orders. In case of high bot activity, we attempted to define who ordered such an activity.

Why this is important?

The Ukrainian population, when receiving information from social media, rarely distinguishes between information from qualified people and propagandists/bots that promote certain narratives.

To make such an important choice for the entire population as vaccination (60-70% of people need to be vaccinated to create collective immunity), a conscious and high-quality assessment of all arguments is necessary. Unfortunately, at the time of the survey, only 47% of Ukrainians were ready to be vaccinated.

Research methodology

The study included posts and comments to the posts on Facebook pages and accounts:

  • 50 largest Ukrainian media outlets;
  • top 1000 Ukrainian FB accounts (by number of subscribers);
  • top 1000 Ukrainian FB pages (Ukrainian pages may also include Russian pages with a large share of Ukrainian subscribers).

Research period: from 01.01.2021 to 13.03.2021.

From all the posts and publications, we identified those that referred to the vaccine or vaccination. To do this, we built an algorithm for classifying posts and analyzed the comments under them. For the sentiment analysis, the algorithm selected comments that had a clearly expressed attitude towards vaccination.

Results of the study

By the number of posts:

  • more than 95% of commenters are people;
  • and yet 7700 bots actively wrote about vaccination under the posts of 2050 largest accounts and pages;
  • each bot leaves 9.4 comments on average, while a human leaves 1.9 comments (which means that every 5th comment on a post is left by a bot);
  • every fifth comment to posts about vaccines is written by bots;
  • among people, the leader in terms of the number of comments - 378 comments to posts, and the most productive bot among the bots wrote 2118 comments.

By the essence of the posts:

  • the percentage of explicitly negative comments from bots is 68%, while from people it is 48%;
  • bots have only 6% of positive posts, while regular people have 10%;
  • many people tend to engage in discussions and discuss other people's opinions in a neutral tone, and bots, performing a specific task, mostly have a clearly defined tone of posts

By the source of control:

  • these are not political bots - we have not found any evidence of their participation in political campaigns;
  • we couldn't find a clear sign of a centralized campaign when many bots use exactly the same comments;
  • according to one version, some of the bots were controlled by russia.

An algorithm to detect an anti-vaccination bot:

This model used three types of data: account profile, account feed, and the account's previous history of commenting on other pages.

  • ommenting friends. Bots often comment together on the same posts. Humans rarely do this. In our sample, 0.15% of people shared the same posts, while 4.2% of all bots did so. The difference is 28 times;
  • bots have a very active and low-quality feed. Bots have an average of 72 posts, while humans have 39. At the same time, 58% of bots' feeds are reposts, while humans have 23%. Bots have an average of 7.3 likes and 1.2 comments per post, while humans have 39.3 likes and six comments;
  • bots have an average account fill rate of 32%, while humans have an average fill rate of 60%. Bots often do not specify gender, city, or birthday;
  • the average number of avatars for bots is 6.8, for people - 18.7. At the same time, half of bots have less than three avatars;
  • bots are also less active in filling out albums. Bots have an average of seven photos, while people have 29;
  • bots rarely make check-ins. Bots have an average of 2.2 check-ins, while humans have 29 (among those who have made at least one check-in).


1. Distribute information among the population of Ukraine about the main differences between bots and people's comments;
2. Disseminate information to the population of Ukraine on the methodology for checking data for disinformation and anti-vaccination narratives;
3. Conduct an information campaign to highlight the results of this study and refute the main negative arguments against vaccination;
4. To help disseminate information from the pages of scientists, doctors and other people who are professionally qualified to evaluate vaccines scientifically.


Other projects




Counteracting disinformation
See more