Looking behind the 'Great Firewall'

  • A statistical analysis of apparent censorship on China's Sina Weibo network may reveal more than the deleted messages themselves

By Jason Togyer

When NBC time-delayed U.S. broadcasts of certain competitions in the Summer Games, disgruntled fans turned to social media to follow their favorite sports in real time, leading some analysts to call 2012 the "Twitter Olympics." According to Advertising Age magazine, during the opening ceremonies alone, about 10 million tweets were sent. (Indeed, some of those tweets were from people complaining that NBC itself was live-tweeting opening ceremonies that the network wouldn't show for hours later.)

Meanwhile, in the People's Republic of China, Advertising Age reports that users of the homegrown Sina Weibo social media network sent some 119 million tweets during the opening ceremonies, "making a strong case to call this the Weibo Olympics as well." 

Sina Weibo is similar to Twitter (including a 140-character limit), but also allows users to attach files, as well as a Facebook-like ability to show comment threads. It's not as well known as Twitter among non-Chinese speakers, in part because an English-language version (although announced) isn't yet available. Within China, however, Sina Weibo controls well over half of the country's micro-blogging traffic and its users include prominent international companies and celebrities; outside China, the site is popular throughout Asia and among Chinese speakers in the U.S., including an increasing number of CMU students and faculty.

Like Twitter, Sina Weibo's wide user base makes it a perfect platform to study trends in language. David Bamman, a Ph.D. student in CMU's Language Technologies Institute, opened a Sina Weibo account in 2011 and started using the service as a research tool to study differences in Mandarin and Cantonese dialects. In June and July, when rumors began to circulate that former Chinese president Jiang Zemin had died, Bamman opened a feed to read those messages.

Imagine his reaction a few months later when he re-opened the feed to find that most of the messages were gone, replaced with the ominous note, "Target Weibo Does Not Exist." Immediately, Bamman saw a whole new area for research. 

Along with Noah Smith, associate professor in the LTI, and Brendan O'Connor, a Ph.D. student in machine learning, Bamman downloaded 56 million messages from Sina Weibo, then sampled 1.3 million messages three months later to see which ones no longer existed. The resulting study by Smith, Bamman and O'Connor, published in March, constitutes one of the first statistical analyses of the results of the Chinese government's attempts to censor or shape public opinion using social media.

"All companies have some sort of internal controls," O'Connor points out. Indeed, U.S. Internet providers routinely clamp down on copyright violations, piracy and trafficking in illegal material. Some companies censor for self-serving reasons; Twitter was roundly attacked for temporary closing the account of a journalist who was sharply critical of NBC's coverage of the Summer Games. 

But speech laws and regulations in China are such that Sina Weibo and similar services inside the country are subject to political censorship as well. Facebook, Twitter and other Western social media sites have been banned from China since July 2009, when the services were used to distribute photos and first-person accounts of rioting in the northwestern city of Urumqi. The Chinese blockade of Western social media has been dubbed "the Great Firewall of China." Time magazine reports that Chinese search engines won't return links to content on sensitive subjects such as the 1989 Tiananmen Square protests or Tibet's bids for independence. 

Sina Weibo is a Chinese counterpart to Twitter. Founded in 2009 by Shanghai-based SINA Corp., Sina Weibo (the word means "microblog" in Mandarin) had more than 300 million registered users as of February 2012, or about 30 percent of all Internet users on the Chinese mainland. Outside observers have assumed that Sina Weibo was either self-censoring or submitting to government censorship of sensitive topics. To test that theory, in December 2010, New York Times editorial columnist Nicholas Kristof opened a Sina Weibo account and in his very first Weibo, asked, "Can we talk about Falun Gong?" Falun Gong is an underground religious movement officially banned in China. Kristof's account was shut down within an hour.

Smith, Bamman and O'Connor found that Weibos about certain sensitive topics, such as Falun Gong, were deleted almost--but not quite--uniformly. Weibos about former Chinese leaders or political exiles were usually deleted as well. Since Weibos contain geographical tags, the survey team was able to determine that Weibos from Tibet were deleted at a much higher rate than Weibos from other regions; about 50 percent of Weibos from Tibet users were deleted, versus about 16 percent for all Weibo users.

But a few topics that rated deletion by Chinese censors surprised the researchers. Following the nuclear accident at Fukushima in Japan, rumors spread around the world, including on the U.S. West Coast, that iodized salt would prevent radiation poisoning. Actually, iodine tablets, not iodized salt, are prescribed to protect the thyroid glands from being damaged by radioactive fallout. 

Bamman, O'Connor and Smith found that following the Fukushima crisis, a high percentage of Weibos about iodized salt were deleted. That could be seen as the government's attempt to make sure people weren't spreading gossip and inaccurate public health information, but in the U.S. and other Western-style democracies, public health officials simply put out press releases to combat the rumors; they didn't resort to censorship.

In days past, totalitarian governments suppressed dissent by controlling printing presses, photocopy machines, radio transmitters--devices that were expensive, difficult to transport or required special skills to operate. Wireless mobile devices are pervasive and social media is easy to use, which makes trying to control microblogging much like trying to stop the ocean's tides.

Yet technology is a double-edged sword. While millions of microbloggers are harder to shut down than a single printing press, digital fingerprints are also easily traceable, O'Connor points out. Sina has added its own new layer of monitoring; it now forcing users to register with a real name, and a "demerit" system allows customers who send out "false" Weibos to be put on warning or blocked. Both the ease of tracking Weibos and the real-name requirement could be expected to have a chilling effect on speech.

As the authors point out, not every Weibo about Falun Gong or Tibet is removed, as would happen under an automatic blacklist. That suggests that human censors are in large part responsible for deciding which Weibos to remove. "I think it might reflect the tension inherent in Sina's role," Bamman says. "It has two different groups of people to satisfy. It's got to satisfy the censors, but it also has to satisfy 365 million users."

What censors allowed to remain on Sina Weibo may say as much about the political climate as what those censors deleted. "It has the potential to expose what the Chinese government thinks is sensitive at a level we didn't have before," Bamman says. During the Soviet era, so-called Kremlinologists studied photos of parades to deduce which Russian leaders were out of favor. Like the Kremlin, China's political leadership makes decisions in secret, without public deliberation. Instead of subjective analysis of artifacts such as photos, students of present-day China might be able to understand the political process better through statistical analysis of social media such as Sina Weibo.

--Jason Togyer (DC '96) is editor of The Link.


For More Information: 
Jason Togyer | 412-268-8721 | jt3y@cs.cmu.edu