By Reuters and Annie Palmer For Dailymail.com
Published: 19:01 GMT, 21 January 2019 | Updated: 20:43 GMT, 21 January 2019
View
comments
A new study has found that social media sites like Facebook and Twitter can learn a shocking amount of information about users, even if they don't have an account.
Researchers from the University of Vermont discovered that these platforms only need access to eight of your one-time contacts in order to infer information about you.
It comes as Silicon Valley giants face increased scrutiny about their data collection practices and whether users have enough control over their private information.
Scroll down for video
A new study has found that social media sites like Facebook and Twitter can learn a shocking amount of information about users, even if they don't have an account
Researchers analyzed data from the public posts of nearly 14,000 Twitter users.
They found that machine learning algorithms could infer with up to 64 percent accuracy what word a user was likely to write next, based on what he and their friends most recently wrote.
That number dropped to 61 percent if only fed with information posted by friends.
If a user doesn't have an account, the algorithm can draw information from up to 8 or 9 of an individual's contacts to predict the user's behavior, the study found.
Although the study focused on Twitter, the same information could be gathered form posts on other social media, like Facebook, provided access to them.
'You alone don't control your privacy on social media platforms,' said Jim Bagrow, a mathematician at the University of Vermont who led the research published in the journal Nature Human Behavior.
'Your friends have a say too.'
Bagrow and his team used statistical models to analyse data from more than 30 million publicly available Twitter posts by almost 14,000 users.
Although the study focused on Twitter, the same information could be gathered form posts on other social media, like Facebook, provided access to them, Bagrow said.
They found that machine learning algorithms may be able to infer with up to 64 percent accuracy what word a user was most likely to write next, based on what he and the people he interacted most often with had previously published.
Accuracy levels dropped only three percent to 61 percent when the algorithms were fed with text posted only by friends, according to the study.
If a user doesn't have an account, the algorithm can draw information from up to 8 or 9 of an individual's contacts to predict the user's behavior, the study found.
The
