Communications of the ACM
Software Extracts Your Location on Twitter Even When It's Secret
Users of online social media reveal more about themselves than they realize, and participation involves a great deal more trust than they think, suggests new research from computer scientists at Northwestern University and the Xerox Palo Alto Research Center.
About one-third of Twitter users do not disclose their location, and in addition to leaving the field blank, many enter inaccurate or even threatening information. Nonetheless, such attempts to maintain privacy from advertisers might not work because the researchers were able to track down Twitter users through subtle hints in their communications with others.
The researchers developed a machine-learning algorithm and applied it to the recent tweets of 10,000 active Twitter users. Although the team was not able to identify the Twitter users by their address or their zip code, they were able to determine the country and state where the users lived.
Moreover, analyzing the data after the fact revealed some terms that were highly predictive of location, such as people who often used the word "Colorado" were from that state. Less intuitive results showed that "biggbi" was highly predictive of residence in Michigan, people who used "gamecock" were likely to be in South Carolina, and people outside Louisiana were less likely to use "crawfish."
From Technology Review
View Full Article
No entries found