The American scientific journal PNAS published a study stating that Twitter algorithms are promoting right-wing politicians. The study was conducted on the tweets of 3.5 thousand politicians from seven countries.
Twitter has experienced that the machine learning algorithms used in the recommender system are biased and produce a result based on data from one specific group of users and ignore other data. To understand the effect of recommendations, Twitter commissioned a study from scientists from leading universities in the US and the UK.
The scientific work considers a sample of more than 46 million unique users, which is 4% of all users of the platform who were influenced by algorithms. In addition to the main participants, the study also included a control group of 11.5 million users who did not see the recommended posts in their feed. The materials for the study were tweets of more than 3 thousand politicians from the USA, Japan, France, Germany, Canada, Spain and the UK. The purpose of the scientific work was to find out the effect of "algorithmic amplification" in the positions of politicians.
The analysis was based on the attention retention effect — if at least 50% of the area of the tweet lingered on the user’s screen for at least 0.5 seconds, then the user is considered to have read the post.
The study gave the following results:
algorithms actually promote political posts, the effect does not depend on which party published the post and whether it is currently in power;
the recommender system does not take into account the political views of users, so the group effect does not develop into an individual one;
posts from the right-wing politicians and parties receive more “algorithmic amplification” than entries from the left, a trend that can be traced in all the countries studied, except Germany;
Canada turned out to be the most contrasting country, where liberal tweets were boosted by 43%, and conservative posts by 167%;
ultra-left and ultra-right parties and politicians do not receive significant strengthening, even if the country's ruling apparatus includes a large number of representatives of these movements;
t has not been found that the recommender system in all countries equally promotes only the ruling parties and ignores the opposition or does otherwise. For example, in the UK, the ruling Conservative Party received the greatest gain, while in Canada, the opposition party.
The researchers decided to go beyond Twitter posts, and analyzed how the algorithms promote political news stories. For the analysis, articles from American publications were used. The result was similar - content from the right-wing media gets more reinforcement, while far-left and far-right sources are hardly promoted.
bbabo.Net