Twitter algorithms amplify right-wing politics

Embargoed until: Publicly released: 2021-12-23 08:56

International

Photo by souvik banerjee on Unsplash

Twitter’s machine learning algorithms amplify messages from right-wing political parties significantly more than the political left, according to new research. In a massive study spanning seven countries, two million active users were released from the ‘personalisation’ model and compared with accounts that remained under the algorithmic influence. Contrary to popular belief, the study found no evidence that far-right and far-left political messaging was amplified more than mainstream political voices. In the US, Twitter’s personalised algorithms promoted right-leaning news media more often than left-leaning outlets. The study was led by Twitter’s own Machine Learning Ethics, Transparency, and Accountability Team.

Media release

From:

Attachments

Note: Not all attachments are visible to the general public. Research URLs will go live after the embargo ends.

Research PNAS, Web page

Editorial / Opinion PNAS, Web page

Supplementary Information PNAS, Web page Link to supporting information

Journal/
conference: PNAS

Research:Paper

Organisation/s: Twitter, University of Cambridge, University College London, University of California, Berkeley

Funder: Funding not specified. Competing interest statement: F.H., S.I.K., C.O., L.B., and A.S. are or were employed by Twitter while this work was carried out. F.H. was a full-time employee of Twitter when the work was started, but left Twitter in July 2020 and continued being affiliated with Twitter through a paid part-time consulting relationship. M.H. was a paid consultant for Twitter. C.O., L.B., and A.S. have a financial interest in Twitter.

Media Contact/s

Contact details are only visible to registered journalists.