LRCCS Spotlight: Blake Miller

Blake Miller Political Science PhD Candidate Statistics MA Student LRCCS Doctoral Fellow

Blake Miller

Political Science PhD Candidate
Statistics MA Student
LRCCS Doctoral Fellow

In this edition of the LRCCS Spotlight series, I sat down with Blake Miller, LRCCS Doctoral Fellow.  In our interview, Miller discusses his research on the fascinating "50-cent party (五毛党)" of government appointed "Internet commentators" and their role in guiding public opinion.

Interview conducted and edited by Eric Couillard.  Transcribed by Erzhan Xu.

Couillard: So tell me about your background. Where do you call home? 

Miller: I am originally from San Diego. But before my family moved to San Diego, I spent a few years of my childhood in Singapore. So a lot of my family friends are Chinese Singaporeans.  I grew up pretty close to Chinese culture and celebrated the Spring Festival with them when I was in San Diego.  I went to Stanford, majored in political science. I was originally interested in American politics, and I took a course on Chinese Politics by Alice Miller. 

After that I became really interested in the Chinese political system, and became really fascinated with Chinese media and how Chinese citizens use the Internet after working for several years as a research assistant with Larry Diamond.  I focused more and more on media, and information control. A lot of the work I do now is on censorship, propaganda, and the 50-cent party.  After I graduated from Stanford, I worked at three different companies as a software engineer. 

Couillard: You were a software engineer? But you studied politics? 

Miller: This is just after I graduated. I had to pay off my student loans somehow! I always knew working in tech wasn’t exactly what I wanted to do, so I came to Michigan, and I have been working with Mary Gallagher on my political science PhD. 

Couillard: You wrote a paper on detecting 50-cent commentators. How do you know when a post comes from the 50-cent party? 

Miller: The paper you are referring to is a method for automatically detecting 50-cent commentators using machine learning and natural language processing tools.  The method identifies 50 cent party members using metadata like IP address, post volume, and social media data. Using a leaked dataset for validation, my model predicts real 50-cent posts from a small district propaganda department with 94% accuracy.

Couillard: Woah, that’s pretty impressive. So who are these people? 

Miller: There is a common misconception based off their name – that they’re supposedly paid 50 cents per post and work part-time.  There is no evidence to support this.  These commentators come from a wide variety of different bureaucracies.  So my research assistants and I have been crawling official government documents for publicly available rosters of “opinion guidance” employees. These rosters and other documents surprisingly are openly available on government websites.  Their most common official title is “internet commentator” (网络评论员). 

Couillard: So do they do this full time? Or is it just part of their job? 

Miller: There are some full-time employees with the title “internet commentator.” During what’s called a ‘public opinion emergency,’ they will rally other people within the bureaucracy to join the core team. These are the only “part-time” commentators to speak of. If there is a public opinion emergency that they can anticipate, they sometimes will have a detailed public opinion contingency plan.  An example is a coal mine collapse. They have plans ready for when one happens and have specific instructions for how commentators are supposed to respond to negative opinion. 

Couillard: So who’s pulling the strings?  Is this coming from the central government or local government?

Miller: “Opinion guidance” teams can be found at several bureaucracies, even prison systems and forestry bureaucracies. Based on some initial findings, it seems like propaganda organs – Internet information offices, propaganda departments, civilization offices, anything that is involved in propaganda – only comprise about 25% of the commentators in my database.

It’s very decentralized, and the tactics and the method of guiding opinion vary a lot.  For example, on an article about the South China Sea tribunal ruling, there are some comments that seemed to echo the party line – deescalate, try to prevent protest, try to keep everything calm, and push faith in the central government’s ability to deal with this problem.  There are a lot of those posts. But there are a lot of other posts where the 50-cent party is attacking individuals, calling people idiots, using very colorful language, calling for war with the U.S., and even flinging racial slurs about Filipinos.

Couillard: So it’s not exactly one voice.

Miller: Not always.  But at the same time, you have things like the Tianjin Port explosion. It seems like in that case, commentators are pretty uniform in their message.  There were lots of concerns in the comments about the government responsibility and corruption.  And the commentators, instead of discussing any of that, shifted the conversation toward talking about how brave the firefighters and first responders were, how sad it was that many died, etc. 

[Editor's note - Click here to read Miller's case studies on the TIanjin port explosion and the South China Sea, co-authored by Mary Gallagher]

From some deep diving into opinion guidance manuals, it seems lots of bureaucracies have plans that are being continually revised.  When something like this happens, they write up reports on how to improve the plan, how they responded to the crisis, how they could have done better. And these 50-cent party members are learning how to better guide opinion, they are like an underground force that is digesting public opinion, responding to it, and in a very iterative way, learning how to best deal with ‘harmful information’, ‘harmful content’.  There is an element of authoritarian learning that the 50-cent party facilitates.

Couillard: That is fascinating and terrifying. So it sounds like what you are saying is they are getting better at their jobs. Do you think they’re successful?

Miller: Whether or not they are actually effective in guiding public opinion or changing minds, is something that I need to spend more time exploring and is actually a major question in my dissertation. I don’t have an answer for that yet. 

Couillard: Can normal internet users sniff out 50 cent people?  Or do they go undetected? 

Miller: I think there is a range in the level of sophistication within a lot of these 50-cent party members.  Some of them, you immediately know can’t be a real person.  Like someone just writing ten messages in a row about how much they love Xi Jinping. It doesn’t appear to be that convincing.  But I mean, the firefighter comments, I don’t think people see those comments and think, “Okay, it looks like I’m being manipulated.”  

At the same time, I’ve found that almost one in five comments that the Chinese news consumer sees is fabricated by the government. So even if some of these comments are detectable to ordinary people, they still might be persuasive in the aggregate.

Couillard: Phew. So what kind of impact would you like for your research to have?  What’s your goal in exploring all this?

Miller:  I think the point of my research is to clarify that online information control in China is not very well understood, and that China's aim to make information control less visible is about getting more accurate information about public opinion. Rather than directly and visibly intervening and creating a chilling effect, the government’s seems to favor guiding opinion from behind the scenes. Without anyone knowing the government is manipulating the information individuals see, they feel free to express their true beliefs, valuable data for the regime. These data, collected en masse, are processed to inform policies and targeted repression.

Thanks for reading!  Stay tuned for more interviews from the LRCCS community!