A study developed by the interfaces group at the Federal University of São Carlos (UFSCar) observed that social media growth grew 821% in the last presidential election, in 2022, pointing to recurring patterns of hate speech.
The research quantified, with computational methods, the phenomenon of regional prejudice on social networks, analyzing 282 million tweets posted between July and December 2022.
The study revealed that, as the election approached, pejorative terms began to appear increasingly associated with the word “northeast” in Twitter posts (currently X). In the month of October, when the two rounds of the election took place, the proportion of posts mentioning the Northeast tripled compared to previous months.
The work was funded by Fapesp and was published in the scientific journal GEMInIS.
Patterns ‘invisible’ to the human eye
To reach these conclusions, the research team used advanced natural language processing (NLP) techniques, a branch of artificial intelligence that allows computers to understand and analyze large volumes of text.
In total, the researchers employed four complementary computational techniques. The main one was the Word2Vec algorithm, which works as a proximity mapper between words.
“This algorithm classifies an association between two terms as 0% [nenhuma associação] a 100% [significado semântico igual]”, explained the authors of the article to the Interfaces press office.
The method allows you to identify which words frequently appear together in the same context. If someone regularly writes “nordestino” next to “ungrateful”, for example, the algorithm detects the association and quantifies it. It is a neutral technique that only maps existing patterns in texts, without making judgments about their content.
The data showed a clear temporal progression: in July 2022, the beginning of the period analyzed, neutral or geographic words predominated in mentions of the Northeast: terms such as “sertão”, “interior” and names of States.
In September, the word “poor” jumped from an association of 57% to 67% with “northeastern”. In October, the month of , the words “ungrateful” (64% association) and “illiterate” (59%) appeared for the first time.
An important methodological detail is that the pejorative words were not pre-selected by the researchers, but emerged from the computational analysis of the data. “This is evidence that the texts collected under the conditions of this research contain sentences that associate Northeasterners and the Brazilian Northeast with ideas associated with such keywords,” stated the scientists.
Historical pattern
The data collected by the researchers corroborates statistics from the NGO Safernet, which operates in cooperation with the Federal Public Ministry. According to the organization, 2022 was the third consecutive election year with significant growth in online hate crimes. In addition to xenophobia, which led the ranking with an increase of 821%, cases of religious intolerance (522%) and misogyny (184%) also increased.
“Regional prejudice in Brazil, specifically against people from the Northeast, constitutes a form of modern xenophobia,” the researchers observed. They explain that these discourses are anchored in historical stereotypes that date back to the end of the 19th century, related to climate, economic and migration issues.
From a legal point of view, the researchers assess that the application of Law 7,716/1989 (Anti-Racism Law) to cases of regional xenophobia remains limited in Brazil. The study provides quantitative evidence that can support discussions about the regulation of digital platforms and content moderation policies.
The researchers point out that digital platforms have technology to detect problematic content – similar to that used to identify copyright violations –, but its application to combat hate speech depends on corporate and regulatory decisions.
