The Terms China’s Twitter Doesn’t Want You To Search

Sina Weibo is China’s Twitter-like service. Photo: jonrussell

If you’re one of the 250 million users of the Chinese social network Weibo and you search for the terms “candle wax,” “evolution,” the color “yellow,” or “China’s military,” you’re going to be disappointed.

These are the findings of a new study and the website Blocked On Weibo, which looks at the search terms blocked on China’s most-popular social network and gives further insight into the mechanics and logic of Chinese censorship.

The author of the project, Jason Q. Ng explains:

For several months last year, I set out to track what one Chinese Internet company, Sina Weibo — China’s leading Twitter copycat — considered off-limits. Utilizing a computer script and much patience, I was able to uncover roughly a thousand unique banned words.

According to that list, I can tell you that Taiwan is mostly fine so long as you’re not discussing Taiwanese independence (台湾独立, 一中一台, etc); all discussion of major religions is allowed except for one, Islam (伊斯兰); and even today, in an age of increasingly open sexuality in China, searching for posts on aphrodisiacs (春药) will return error messages.

Many of the politically sensitive terms, such as Dalai Lama or “Mein Kampf,” are unsurprising. Other terms — seemingly innocuous to non-Chinese speakers — are banned because of their double meanings. For example, “yellow” is banned because it can describe pornography and “candle wax” refers to a sexual practice (don’t ask).

With Facebook and Twitter blocked in China, Chinese users have flocked to social networks such as Weibo, Tencent, and Sohu. Social networks have been a way for citizens to get around state censorship and, in some cases, criticize public officials. After the July 2011 high-speed train crash that left 40 people dead, there was an unprecedented outpouring of anger on Weibo.

China’s “Great Firewall,” rather than being a monolithic structure is, in reality, a complex lattice of infrastructure, hardware, software, and legal regulation. Part of that lattice filters IP addresses for certain websites; another part screws with the domain name system so if you type in a web address it will either be unresolved or take you somewhere else.

Given the size of the Chinese Internet, China outsources much of its censorship work to its tech companies. The government encourages companies like Weibo to self-censor by releasing vague, but ominous, government directives. (For example, in an official 2010 White Paper on Internet use, problematic content is described as “propagating heretical or superstitious ideas; spreading rumors, disrupting social order and stability.”) As Ng writes, “the fact that there is no officially published blacklist, coupled with the fear of severe punishments, compels them to step even farther back from the imaginary line.“

The main method of preventing content from going viral on social networks is blocking search terms for sensitive subjects. But companies like Weibo also delete messages as well.

A new study by computer scientists at Carnegie Mellon University in Pittsburgh gives more insight into which terms are deleted on Weibo. The study, which draws on a statistical analysis of 56 million messages on Weibo between 27 June and 30 September 2011, claims to be “the first large-scale analysis of political content censorship in social media” in China.

Like the banned search terms, the deleted messages were mostly unsurprising:

Several interesting categories emerge. One is the clear presence of known politically sensitive terms, such as 方滨兴 (Fang Binxing, the architect of the GFW), 真理部 (“Ministry of Truth,” a reference to state propaganda), and 法轮功 (Falun Gong, a banned spiritual group). Another is a set of terms that appear to have become sensitive due to changing real–world events. One example of this is the term 请辞 (to ask someone to resign); deleted messages containing this term call for the resignation of Sheng Guangzu, the Minister of Railways, following the Wenzhou train crash in July (Chan and Duncan, 2011). Another example is the term 两会 (two meetings): this term primarily denotes the joint annual meeting of the National People’s Congress and the Chinese People’s Political Consultative Conference, but also emerged as a code word for “planned protest” during the suppression of pro-democracy rallies in February and March 2011 (Kent, 2011).

Other deletions show the paternalistic side of the Communist Party in preventing panic, for instance quashing rumors that iodized salt protected against radiation poisoning after the Fukushima disaster. There was a direct government order to quash this particular rumor and Weibo reportedly has a 10-person “rumor control” team.

But are the deletions just based on a message’s content? Or might they depend on the number of times the message was rebroadcast or by how influential the author was? The study found that the differences between deletion rates of messages that were widely rebroadcast against those that were not was statistically insignificant. Nor, in the study’s sample, did the number of followers seem to determine whether a message would be deleted.

What the authors did find was the importance of geographic location. By looking at location-related metadata attached to accounts, the study found that in trouble spots such as Tibet there was a deletion rate of 53 percent; in Beijing, the rate was 12 percent:

We might suspect that higher rates of deletion in these areas may be connected with their histories of unrest (especially in Tibet, Qinghai, Gansu, and Xinjiang), but there are several possible alternative explanations: Sina censors may be deleting messages with greater frequency due to increased attention to these areas, perhaps enabled by the comparatively small volume of messages originating from them.

For Weibo, filtering and deletions are clearly a high priority to keep the Chinese government sweet. As the Carnegie Mellon report points out, the Weibo CEO has said that the company employs at least 100 censors, although “that figure is thought to be a low estimate.”

On the one hand Weibo’s censorship system is agile and responsive, by focusing on trouble spots and manually stepping in to nip rumors in the bud. But it can also be clunky. For example, Ng points out that the gaming console Nintendo 64 is a blocked search term as “64” is “short for June 4, the day of the 1989 Tiananmen crackdown.” And the Carnegie Mellon study showed that positive messages about a certain sensitive subject were also deleted, although it was inconclusive as to why that was.

The Chinese government is increasingly wary of social media as an unregulated space, not just in terms of free speech but also spammers and illicit trade. A new regulation, [that came into force in March 2012, means that all users of Chinese microblogs had to verify their accounts with their official IDs.

This article was written by Luke Allnutt and was originally published by Radio Free Europe on March 21, 2012.