China's social networks hit by censorship, says study

China steps up censorship of banned terms in regions where unrest was high

Chinese censors are actively targeting social media to quash discussion of banned topics, suggests research.

The US study gives the most in-depth look at the extent of China's policing of discussions on microblogging sites.

Analysis of almost 60 million messages from China's equivalent of Twitter suggested which topics were banned.

It also revealed that China tuned its censoring activity to be more aggressive in places where political unrest was high.

Word search

The study, reported in New Scientist, by David Bamman, Brendan O'Connor and Prof Noah Smith from the Language Technologies Institute at Carnegie Mellon analysed short messages sent via the Sina Weibo service.

The public programming interface to the Sina site let the trio grab 57 million messages sent between June 27 and September 30, 2011. Three months later they checked to see which messages disappeared from the service to identify which terms caught the attention of the censors.

The work showed that the social media censor was similar to the system overseeing Chinese web access.

That system, known as the Great Firewall, stops people visiting some sites outside China, returns no results for searches of banned terms, censors chat and vets blogs. Banned topics include the Falun Gong spiritual movement and human rights activist Ai Weiwei.

In a similar way, the study found that messages containing these banned terms tended to be deleted from the Sina Weibo service.

The censorship system governing social media sites such as Sina Weibo is not foolproof

It also found that the censorship system could be quite nimble and react quickly when words or phrases start to assume a more political meaning. For instance, the word "lianghui" became sensitive when it started to be used as a code word for a "planned protest".

Similarly, a word meaning "asking someone to resign" became sensitive in the wake of the high speed train crash in July 2011 that killed 40 people. Mistakes by officials have been blamed for leading to the disaster.

The study also found significant variation in how active the system was in different regions of China. In Tibet about 50% of messages were deleted, compared to 12% in Beijing and 11% in Shanghai.

Mr Bamman said he was surprised at the extent of the censorship and at the fact that some banned terms, such as Falun Gong, were appearing at all.

"The fact that we see these terms in messages would seem to imply that the censorship is not an automatic process," he told the BBC. This exposed, he said, the tension between government demands for active policing and reluctance on the part of companies to inhibit what their customers do.

He speculated that the sheer amount of messages passing through the social media services might also make it harder to censor all the content the government finds troubling.

And, he added, many people would find ways around the controls.

"People will talk about what they want to talk about," he said, "they may just have to find different ways to say it."

A paper detailing the study is due to be published in the March edition of the First Monday journal.