China: Researchers Uncover Microblog Filtering Mechanisms

After analyzing over 490 million posted messages from users of Sina Weibo, China's most popular microblogging service, a group of US-based computer scientists are beginning to understand how the platform's censorship mechanisms work.

In a research paper entitled “The Velocity of Censorship: High-Fidelity Detection of Microblog Post Deletions,” the group describes how they collected Sina Weibo posts both from common and “dissident” Weibo users. Launched in 2010, Sina Weibo had over 500 million users by end of 2012. Over 46 million users post messages on a daily basis.

Researchers used messages both from Sina Weibo's public timeline and from a roster of over 3500 “dissident” users of the service. Sina Weibo aggregates all user messages in a publicly accessible, reverse chronological list, known as a public timeline — unless those messages have been censored. In order to understand more about censorship on Sina Weibo, researchers also studied messages sent by users who had been cited by watchdog news site China Digital Times and those who had had 5 or more posts deleted in the past. Among this group of “dissident” users, the researchers found that an average of 12.75% of messages are deleted. Of these, 82.25% are “child” posts (comments and reposts of an original post.) “Support Syrian rebels,” “Beijing rainstorms,” “human rights news,” “group sex” were among commonly censored terms.

In mainland China, public commentary that is interpreted as a challenge to the rule of the Chinese government or China's Communist Party is considered politically sensitive and thus, by government and party standards, should be censored. Yet a mechanical filtering system that targets terms such as “government” is both unrealistic and useless. Using the term “government” as an example, netizens could easily substitute the term with “ZF” [the Chinese pingyin or transliteration of “Zhenfu,” which means “government”] or other terms such as “empire” when they want to criticize authorities.

The censorship body has developed a very sophisticated system and hired a large number of site moderators to monitor and, when necessary, stop the flow of dissenting ideas on the social media platform. As Sina Weibo is the most influential social media platform in China, many believe that it is thus placed under the heaviest degree of scrutiny and is more subject to censorship than other platforms.

During the research period, Sina Weibo users posted an average of 70,000 messages per minute. The study found that 5% of deletions took place 5-8 minutes after a message was posted; 30%  of deletions were made within 30 minutes of posting; and 90% within one day. So how does the Sina Weibo manage to censor such huge quantities of posts so quickly? Weibo moderators use several methods.

Keyword filtering

The most commonly known method is the use of filters for specific terms or keywords. Deletion data showed that the Weibo censorship system maintains multiple keyword lists, with each list triggering a different kind of response. Different kinds of keywords can trigger explicit filtering, implicit filtering, or “camouflaging”.

Explicit Filtering: When users send a message containing a keyword that is marked for explicit filtering, they typically receive a notice saying that the content of the post is forbidden by law.

For example, when you type in Chinese President Hu Jintao's name, a notice will appear saying that your post has violated either Weibo community rules or existing law. However, it is important to note that Hu Jintao's name is searchable, which means certain user accounts have the rights to post this particular sensitive word.

Warning message issued by the Weibo system when the post contains the word "Hu Jintao" in Chinese.

Warning message issued by the Weibo system when the post contains the word “Hu Jintao” in Chinese.

Implicit Filtering: Soon after users send a message, they receive a notice telling them that the post will be reviewed and the published with a time delay. This kind of keyword filtering usually targets specific users or a particular hot topic.

Camouflaged Posts: The published message is hidden from other Weibo users — only the message's author can see it.

Apart from the above three responses, the public timeline and search results also have either content or user filters that block certain content from appearing.

User monitoring

The system also maintains lists of users whose messages are considered politically sensitive. Similar to keyword filters, each user group is subject to a different level of monitoring. Some users undergo routine deletion; some cannot upload messages without having them reviewed by the Weibo moderators (implicit filtering); some have their accounts temporarily suspended. The level of monitoring may change, depending on the issue in question. For example, every year when the commemoration of the June 4 Tianenmen Square protests is approaching, messages from Hong Kong IP addresses are closely monitored. Users who try to commemorate June 4 in Weibo typically have their accounts suspended for a few days.

Deletion patterns

The deletion of a message has a chain effect. When a message is deleted, its re-posts will also be deleted automatically. In addition, the moderators will also search the newly identified sensitive term in its system archive to make sure that previous mentions of the term are eradicated.

The censorship body watches hot topics (similar to trending topics on Twitter) closely. When a hot topic is politicized, or involves sensitive themes (such as government, political reform, etc.), massive removal happens fastest.

As the research results are mainly derived from the group of dissident users, the speed and rate of deletion does not reflect an ordinary user's experience. Yet the research has provided some insights into how Sina monitors and censors the massive quantity of messages that appear on its platform.

Future studies might benefit from fuller consideration of how the Chinese authorities control information flow online must consider the role of the so-called 50 cent party, a group of Internet commentators who promote government policies and favorable opinions of the government on social media platforms in China.

See original article for further information:
The Velocity of Censorship: High-Fidelity Detection of Mircroblog Post Deletions [pdf]

Additional recent commentary on “The Velocity of Censorship”:

Computer Scientists Measure the Speed of Censorship on China's Twitter

MIT Technology Review

Scientists Measure Speed of Censorship on Sina Weibo

China Digital Times

Censorship Alert! Academic Study Shows Sina Weibo's Human Censors Are Pretty Darn Fast

Tech in Asia

10 comments

Join the conversation

Authors, please log in »

Guidelines

  • All comments are reviewed by a moderator. Do not submit your comment more than once or it may be identified as spam.
  • Please treat others with respect. Comments containing hate speech, obscenity, and personal attacks will not be approved.