Twitter Algorithms Amplify Conservative Politics
SAN ANTONIO — Twitter content curated by its personalization algorithms amplifies the mainstream political right more than the left, according to a joint study conducted by the platform’s transparency and accountability team.
Researchers undertook a large-scale experiment that analyzed millions of Twitter users, political parties in seven countries and 6.2 million news articles shared in the United States to conduct a comprehensive audit of Twitter’s algorithms. In their findings, the researchers concluded that mainstream political figures on the right experienced higher algorithmic amplification than the mainstream left in six of the seven countries studied in the report, including the United States.
The tweets examined in the study originated from the U.S., Japan, the United Kingdom, France, Spain, Canada and Germany. The experiment included a randomized control group consisting of nearly 2 million daily active accounts that were presented without algorithmic personalization, which can be disabled from users’ feeds.
Further, the researchers studied the landscape of news media in the U.S. and found that strong partisan bias in news reporting is associated with higher amplification. The study was conducted by Ferenc Huszár, Sofia Ira Ktena, Conor O’Brien, Luca Belli and Andrew Schlaikjer of Twitter’s Machine Learning Ethics, Transparency and Accountability Team, and Moritz Hardt of the University of California, Berkeley.
“Personalized ranking prioritizes some tweets over others on the basis of content features, social connectivity, and user activity,” the text of the report read. “There is evidence that different political groups use Twitter differently to achieve political goals. What has remained a matter of debate, however, is whether or not any ranking advantage falls along established political contours, such as the left or right, the center or the extremes, specific parties, or news sources of a certain political inclination.”
Prior to the algorithm’s institution, Twitter presented users with content exclusively from accounts they followed in reverse chronological order. In 2016, Twitter began presenting content on users’ feeds by using machine learning algorithms based on a “personal relevance model,” meaning some content is prioritized over other content due to a ranking system based on user preferences.
The researchers quantified the extent of different political groups’ assistance from algorithmic personalization by measuring their respective amplification in the lead up to the experiment. They then constructed an amplification ratio that intersects a set of tweets and a set of users who encountered the tweets on their feeds.
The data collected by the researchers doesn’t suggest the Twitter algorithm amplifies extremist ideologies from either side of the political spectrum over content from moderate users. The amplification of far-left or far-right parties is generally less than that of moderate or centrist parties from the same country.
“Our analysis of far-left and far-right parties in various countries does not support the hypothesis that algorithmic personalization amplifies extreme ideologies more than mainstream political voices,” the text of the report read. “However, some findings point at the possibility that strong partisan bias in news reporting is associated with higher amplification.”
While some recently proposed legislation takes aim at discrimination on social media and limiting protections that prevent online platforms from being held liable for third-party content, no bills that would regulate personalization algorithms have been filed. Both the SAFE TECH Act introduced by Sen. Mark Warner, D-Va., and the Justice Against Malicious Algorithms Act introduced by Rep. Frank Pallone, D-N.J., would amend section 230 of the Communications Act in different ways but neither would force Twitter to alter its personalization algorithm or personalized relevance model.
The SAFE TECH Act would hold social media sites accountable in cases of cyberstalking, targeted harassment and discrimination on their platforms. The Justice Against Malicious Algorithms Act would remove absolute immunity in instances where algorithmic manipulation causes physical or severe emotional injury. Both bills have been introduced in their respective chambers of Congress and await committee hearings.
“Across the seven countries we studied, we found that mainstream right-wing parties benefit at least as much, and often substantially more, from algorithmic personalization than their left-wing counterparts,” the researchers concluded in the study. “In agreement with this, we found that content from U.S. media outlets with a strong right-leaning bias are amplified marginally more than content from left-leaning sources. However, when making comparisons based on the amplification of [an] individual politician’s account, rather than parties in aggregate, we found no association between amplification and party membership.”
Reece can be reached at [email protected]