Despite 300 million users generating more than 100 million posts per day, Chinese censors are very adept at deleting unwanted material on Weibo, the country’s main blogging platform. The censors are also fast,
according to new research.
“What’s interesting is how effective they’re able to be despite growing to a large size,” said Dan Wallach of Rice University, one of the authors of the study.
The research, conducted by computer scientists from Rice University, Bowdoin College and the University of New Mexico, showed that 30 percent of deletions on Weibo occur within five to 30 minutes after posting and that nearly 90 percent of the deletions happen within the first 24 hours. Sina Corp.'s Weibo is a microblogging service similar to Twitter, which is blocked in China.
In order to find how quickly posts were removed, researchers identified about 3,500 Weibo users with histories of deleted posts. These users were carefully monitored from July 20, 2012 to September 8.
Reading the lists of top deleted items from
an earlier version of the paper is almost like reading a newspaper.
For example, on July 22, the number one banned topic was “Beijing rainstorms,” after netizens voiced anger toward the government’s response to storms that left at least 77 dead. The topic remained among the top banned stories for several days.
Then, in late July and early August, Gu Kailai crept into the top banned topic list. Gu Kailai is the wife of disgraced politician Bo Xilai and was convicted of killing a British businessman in one of the biggest scandals to rock China in recent years. She again appeared among the most blocked on August 20, the day she received a suspended death sentence.
Despite growing tensions with Japan over the disputed Senkaku/Diaoyu islands, Weibo keeps tight control over the online conversations. The topic “anti-Japanese” was the most blocked topic on August 17, 19 and 20 after an initial wave of anti-Japanese rioting broke out in China.
Wallach says “there’s no question that a small roomful of humans can’t keep up with 300 million posts without a lot of help from the machine.”
The sheer volume of Weibo posts points to a high level of automation, he added.
According to the report, if an efficient worker could read 50 posts per minute, 1,400 censors would be needed to read the 70,000 posts Weibo generates per minute. If each worker worked an eight-hour shift, it would take over 4,000 workers to delete sensitive posts. That’s clearly not happening said Wallach.
“Based on our data, it sounds like there’s a policy office there and there are people whose job it is to say ‘holy crap, there’s way too much discussion on the North Korean nuclear test. Shut it down,’” he said. Once that policy has been decided, he said, the deletions begin in earnest.
The research also showed that the number of deletions dropped off from midnight until around 4 a.m. and then spiked in the early morning as censors caught up on overnight posts and the flood of early morning posts.
Wallach points out that Weibo is a company, not a government entity, and that in order for it to succeed, it has to stay on good terms with the government.
“You have to do censorship that keeps the government off your back. They’re walking a fine line,” he said, adding that Weibo censorship doesn’t have to be perfect, just good enough.
While some topics such as the one-child policy, freedom of speech and “despise gov” are obvious targets for the censors, Wallach said some off-limit topics were surprising.
“It never would have occurred to me that pornography and gambling are things you’re not supposed to talk about,” he said. “They don’t want you talking about group sex.”
On July 31, just the word “accident” was among the most banned topics.
One way Weibo users try to get around the censors is to use nicknames, code words or anagrams to refer to sensitive issues obliquely. Another trick is to swap one Chinese character that might trigger censorship with another that resembles it.
“There’s all kinds of interesting games that you can play to stay away from an auto-censor,” said Wallach. “Eventually, someone in the policy department figures it out.” After that, he said, it’s a relatively simple process to write some software to detect the substitution and “make it go away.”
Whether that combination of computers and humans can continue to dampen conversations about taboo subjects “is an interesting question,” said Wallach.