Internet researchers have long known that the Chinese government manipulates content on the internet. Not only does it engage in heavy censorship, but it also employs hundreds of thousands of people, the so-called "50 cent army", to write comments on the internet. New research by Gary King, Jennifer Pan and Margaret Roberts (whom I'll refer to as KPR for convenience) uses sophisticated techniques of gathering and analysing massive amounts of data to tell us what is going on.
• The fake commenters are being paid by the government.
It's hard to figure out exactly who is being paid to comment by the Chinese Government and who is not. While accusations of membership in the 50 cent army are rife on social media, these accusations are at best unreliable, and at worst downright misleading. However, KPR were able to take advantage of a major leak of information to figure out what is happening. A blogger released a trove of emails from the propaganda office of a mid-sized unit of China's local government. These emails included a little over 40,000 unambiguous examples of 50 cent army comments. KPR were able to use these comments to draw conclusions about how fake commenting works. They were also able to train a specialised machine learning algorithm to hunt down and identify similar comments on Chinese social media so that they could get a broader understanding of the ecology of government-sponsored comments in China.
One initial result is surprising. Many people think of the 50 cent army as independent contractors who are paid a small amount of money for each comment they make. If KPR are right, then this is far from the mark. The commenters identified in the leak appear to be regular government employees. The sample of emails that King, Pan and Roberts analyse suggest that they work directly for the Communist Party or for different organs of the local government, and presumably are expected to write these comments as part of their official duties.
KPR use statistical techniques to figure out how many social media comments are generated by people paid by the Government. The results are startling. Government employees generate about 448 million comments every year. A little over half of these comments are made on government sites, albeit pretending to be comments made by ordinary citizens. The rest are made on commercial sites, mixed into streams with family news, dog photos and the like. The result, as KPR describe it, is that a "large proportion of government website comments, and about one of every 178 social media posts on commercial sites, are fabricated by the government".
To be clear, these figures depend on a certain amount of extrapolation and educated guesswork, which KPR describe in the paper. Even so, their results are plausible. They asked a random sample of the people whom their techniques identified as paid to write for the Government, whether they were doing this professionally. They also asked care people who they knew to be paid by the Government, because they were identified thanks to the leak, whether or not they were paid professionals. More or less the same percentage of both groups - nearly 60 per cent - admitted that they were.
• Fake commenters are not paid to stir up controversy.
There are many popular rumours about what government-paid commenters do. Some - especially non-Chinese commentators - think they are paid to stir up hatred and resentment of foreign countries such as the United States. Others believe that they are paid to respond to criticism of the government with bogus argumentative talking points.
KPR's evidence suggests that both of these are incorrect. Paid government commenters don't seem to say many nasty things about foreigners. Nor, for that matter, do they engage in argument on the internet. Instead, they praise and distract. They write posts that cheer lead for the Government. They also try to distract the public, especially when they fear that there might be protests or other social and political activity that might be dangerous for the Government. They don't appear to care particularly when people complain about the Government. Instead, they act when there is a real risk of popular upheaval. In KPR's words:
"Since disrupting discussion of grievances only limits information that is otherwise useful to the regime, the leaders have little reason to censor it, argue with it, or flood the net with opposing viewpoints. What is risky for the regime, and therefore vigorously opposed through large scale censorship and huge numbers of fabricated social media posts, is posts with collective action potential."