Users with a high probability of automation, bots were responsible for 12% of posts in Portuguese on Twitter about the Ukrainian War in the second week of the conflict. Analysis of Pegabot, a tool from the Institute of Technology and Society of Rio de Janeiro (ITS-Rio), shows that 15% of accounts that posted about the confrontation between March 2 and 9 have a 70% chance of being robots.
The study analyzed a set of keywords present in the messages —Ukraine, Russia, Kiev, Moscow, Vladimir Putin and Volodymyr Zelensky, including variations in spellings. In the posts of automated users, “Russia” and “Ukraine” were the most cited words, present in 36.2% and 35.2% of the tweets attributed to robots, respectively. Then came “Kiev” (13.8%), “Zelensky” (7.9%), “Moscow” (2.9%) and “Vladimir Putin” (2.1%). The spelling variations did not exceed 1% each.
The volume of automated behavior was above that found in other analyzes in the past.
“Usually the volume of users with automated behavior is between 6% and 9%. This survey was one of those that returned the highest volume of users in this category”, says data scientist Malu Mondelli, from ITS-Rio. “The volume of users who showed high levels of automation and the number of tweets serve as a warning sign that discussions on the subject in Brazil also rely on the use of automation techniques.”
A total of 725,475 tweets were analyzed, published by 13,188 profiles Of this total, 1,979 of the users are more than 70% likely to be automated, according to Pegabot’s analysis criteria.
The accounts were responsible for approximately 84,000 messages within a week. Of these, 71,000 were retweets. The number of posts from these profiles remained similar, ranging from 8% of the total, on March 3, to 13% of the total on the 6th, 7th and 8th, dates when there was the peak of publications.
Another common factor identified was the creation date on Twitter. A graph shows that 75% of these accounts were created in the last five years, with an acceleration from 2018, the date of the last presidential election.
Mondelli states, however, that it was not possible to verify whether the profiles were somehow directing the public debate. “But we could identify automated users interacting more with profiles with right-wing political alignment — @biakicis and @taoquei, for example.”
The three publications most retweeted by automated accounts, for example, criticize the Free Brazil Movement (MBL) for not reporting on money raised to help Ukraine. The tweet most replicated by the robots was by state representative Janaina Paschoal (PRTB-SP).
In second place was a tweet by journalist Kim Paim also demanding accountability from the organization, citing former minister Sergio Moro.
In third place was a message on the same topic from the coordinator of the Brazilian Lawyers Movement, Flavia Ferronato.
Themes of national politics also appear among the posts with the highest engagement published by the robots. The message with the highest volume of interactions criticizes former president Luiz Inácio Lula da Silva (PT): “In Ukraine, prisoners were released to go to war. Here they are released to run for president.”
The second compares the five-day deadline for the Planalto Palace to explain the trip of Rio de Janeiro councilor Carlos Bolsonaro (Republicans) to Russia, given by the Minister of the Federal Supreme Court Alexandre de Moraes, to an inquiry against the PT president. , Gleisi Hoffmann, who would have been stopped for four years in court.
Profile diagnosis
To conclude that a user has a high probability of automated behavior, Pegabot analyzes a set of four criteria: user profile, ie account name, number of followers and followed accounts, description text, number of posts and favorites; frequency and time of publications; network, which consists of checking a sample of the timeline, considering hashtags and mentions to the profile; and sentiment, that is, the positive or negative content of the most recent 100 tweets.
All of this results in a score ranging from 0, which represents no probability of automation, to 100, for a high chance of automated behavior.
Asked by Agência Lupa about the action of robots in messages about the Ukrainian War on the platform, Twitter said in a note that a preliminary analysis of the profiles detected by Pegabot indicates that there is no evidence of automated activity in the proportion indicated in the study.
“It is worth remembering that applications that try to guess whether or not accounts are robots only access external signals from the accounts, which can lead to false positives”, says the text. The platform also stated that it has a policy against spam and manipulation and that it has identified these behaviors and acted to curb them.
