How Censorship in China Allows Government Criticism but Silences Collective Expression (or why locally-led and implemented censorship works in the fractured landscape of Chinese social media)

I am reading articles on technology and public policy because one day I want to do a PhD on this stuff. To make sure I read the articles, I'm summarising what I remember here and filing it under 'professional stuff' in my labels. The parenthetical statements are usually mine while the rest is where I try to accurately represent what I took from the article. 

Title: How Censorship in China Allows Government Criticism but Silences Collective Expression

Authors: @kinggary, Margaret Roberts & Jennifer Pan

Summary

This much-lauded paper gives strong evidence that the Chinese government censorship policy for social media aims at quickly removing any and all content that represents, reinforces or spurs social mobilisation, regardless of the actual subject or tone of the censored comment.

So the Chinese government censors posts that could lead to offline mass assemblies rather than posts that criticise Chinese political policies or leaders.

A bit more

Chinese censorship is amazingly effective, managing to accurately and quickly (usually within 24 hours) censor huge amounts of sensitive social media activity when specific events occur that lead to mass posts online by the Chinese public.

The sole exception to the above appears to be the more consistent and subject-specific censorship of pornography and critical posts referring specifically to the Chinese censors themselves (vs. any other Chinese political / governmental system or leader, which the censors don't mind overmuch as these unkind and sometimes downright rude posts remain largely untouched).


Chinese social media - how it probably works (according to the article)

Unlike the west, where specific social media sites dominate most users (e.g. Facebook, Twitter, Tumblr...), China hosts a devolved mass of social media sites, many of which are local forums (bbs).

The study for the article looked at posts from a wide array of social sites including one massively popular site, blog.sina (originator of 59% of all the posts reviewed), as well as a lot of local sites.

Given the disjointed and localised nature of Chinese social media's infrastructure, guidelines on censorship need to be clear and consistent from Bejing on down.

Chinese censorship - how it probably works (according to the article)

Chinese censorship has three levels covered in the article (and below I add a fourth, also referred to a bit in the article):

1. The Great Firewall of China - This is the infamous firewall by which China excludes entire websites such as Facebook (China instead has RenRen) and Twitter (China has Weibo).
2. Keyword blocking - This is a rudimentary censorship system where the government automatically blocks a post using any of a number of pre-identified keywords. Chinese tend to get around this by using homophones (words that sound similar to what they really want to post) and homographs (words that look similar to what they really want to post).
3. Censors - Employing people to comb through the millions of posts to delete any that are objectionable. The Chinese government reportedly employs 20 000 to 50 000 internet police, or wang jing, and Internet monitors, or wang guanban; 250 000 to 300 000 "50 cent party members" at various level of government, or wumao dang; and up to 1 000 individual censors per Chinese website.
4. Encouragement of self-censorship - Aside from the fact that now you can get jail time in China if your objectionable post is reTweeted (shared) 500 times, people living in any online community tend to be careful about what you post. (In the USA, a 17-day study by Facebook found that 33% of all posts were self-censored, getting the boot before the user mad the post public.) That, combined with the sometimes dangerous results of online rabble-rousing experienced by certain Chinese bloggers, means that a lot of the censoring happens before (or just after) user puts fingers to keyboard. 

The two theories of censorship explored by the article

The article compares two theories of censorship explored in other academic literature related to China:

• State Critique - Does the Chinese State censor any and all criticism of its policies, system, and leaders?
• Collective action potential - Does the Chinese State censor stuff that could galvanise people to get together and protest/discuss/assemble with regard to an issue or an event? (Hint - this is the winning theory in this article - by a looooong shot).

How did the authors run the study to write the article?

1. The authors oversaw a study that used undisclosed automated methods to collect over 11 million posts from the Chinese social web for different time periods. Each time period reviewed lasted 6 months. 
2. Each of these posts were automatically classified according to 85 topics. Each topic fell into one of three categories of sensitivity: 
• "high" e.g. the arrest of the dissident artist Ai Weiwei; 
• "medium" e.g. the one-child policy in China; and 
• "low" e.g. discussion about video games.
3. They coded the posts - using data mining and machine-learning algorithms (to ensure that they didn't need to translate posts, which would lead to inaccuracy). Coding rules considered the following constraints:
• The automatic classification of all posts into 85 topic areas based on pre-defined keywords. (11 million posts is a lot to do by hand - so they got the machines to do it.)
• Since social media is, as the article points out, "bursty" - tending to be a little about nothing for long stretches followed by a lot about one thing for a 2 to 3 day period. Posts were organised around bursts linked to specific offline events.
• Each burst was linked to its offline event (the one thing that was on everyone's finger tips for the 2-3 day "bursty" period). 1 of 5 coding categories were then assigned to each burst/event:
• collective action potential - "the pursuit of goals by more than one person controlled or spurred by actors other than government officials or their agents. 
• criticism of the censors
• pornography
• government policies
• other news
• Two well-educated humans were given a sample of bursty events to code to double-check the machine coding. The machine coding was given a 98.9% accuracy rating by the two humans. 

Some examples of highly censored events

Posts referring to the following events tended to be censored a lot, whether or not the individual posts were positive or negative or neutral with regard to the government:

• Inner Mongolia discussions related to the death of an ethnic Mongolian herder by coal truck drivers
• Posts about an altercation between a pregnant woman and security workers in Zengcheng province
• Posts about Ai Weiwei's arrest
• bombings in Fuzhou over land claims
•  after the Fukushima nuclear explosion in Japan, online rumour starting in Zhejiang that salt could protect people against radiation
• Chen Fai, the environmental activist who successfully established the first "no plastic bag village" in China - Censorship here is particularly telling. Chen Fai has the support of the national government, but, due to his association with collective action, posts about Chen Fai are heavily censored presumably by local government. 
• Posts in a specific province that blamed the Tianneng Group when a battery-making factory allegedly poisoned local children by releasing lead into the water supply. 

So in conclusion...

Chinese censorship, much of it manually-implemented at the local level, aims to dissuade collective action.