The stunning speed of China's Internet censors

Researchers tracked the fate of 2.38 million posts on China's most popular microblogging site, Sina Weibo.
Written by Janet Fang, Contributor

It takes minutes, according to a new study on the mechanics of censorship on China’s most popular microblogging site, Sina Weibo.

Chinese Internet companies police their own content, Wall Street Journal explains, and Sina’s team of "editors" are at the heart of the world’s largest effort to control social media.

The speed is astonishing considering that 70,000 messages are posted per minute on average, and the entire process can't possibly be fully automated.

The study authors, one independent researcher and four American computer scientists, tracked the fate of 2.38 million Weibo posts published between July and September 2012. The findings:

  • Deletions happen most heavily in the first hour
  • 5% of deletions happened in the first 8 minutes, and within 30 minutes, 30% of deletions were finished
  • 90% of deletions happen within the first 24 hours
  • 82% of deleted messages were reposts of earlier messages
  • Reposts are tracked down and deleted within 5 minutes
  • 300 of the 3,500 accounts tracked were deleted when a user repeatedly posted sensitive content

This study also laid out some possible approaches. Censors probably refer to lists of sensitive terms and follow posters likely to say sensitive things. Most deleted posts contain code words to evade automatic keyword filtering; but as censors uncover fresh code words, they can use keyword searches to go back and delete them.

Censorship happens 24-hours a day, but there’s a lull in the early morning hours and when the news comes on at 7 p.m.

Previous work by the same team found that: “starting from the point where censorship begins, the frequency of the topic drops dramatically.”

But as the BBC reports: if Weibo had insufficient controls, the government may take action; if their controls were too rigid, users would abandon them for their competitors.

The work [pdf] is available at arXiv.

[Via WSJ, BBC]

Image by bfishadow via Flickr

This post was originally published on Smartplanet.com

Editorial standards