For the past several years, human rights organizations around the world have kept a close eye—insofar as it has been possible—on human rights abuses against Uyghur Muslims by the Chinese government. It has been reported that in the Xinjiang province of China, Uyghurs are subjected to forced religious “re-education” and labor camps. As with many important political issues, conversation has been driven largely by popular culture events and social media. For example, in 2019, NBA superstar Lebron James came under fire for his business ties with China, which, critics allege prevent him from speaking up for Uyghur people. A year later, Disney’s “Mulan” created an unignorable flashpoint in the U.S.; having been filmed partly in Xinjiang, the company employed several state-aligned businesses in the region to help facilitate film production in China, leading to loud calls for a boycott against the entertainment giant.
Furor over human rights abuses in Xinjiang pose a serious threat to Chinese business interests across the globe. Already, a number of Western clothing brands have become embroiled in controversies of their own over their use of Uyghur cotton, leading some to pull out of the region entirely. The official Chinese government response to these increased Western concerns over the treatment of Uyghur people in Xinjiang has been largely defiant. Officials argue against companies “politicis[ing]” their “economic behavior” and outright deny abusing Uyghur Muslims.
But official outlets are not the only way China has sought to sway global hearts and minds. Understanding the role of public outcry on international perceptions and business decisions, the Chinese government, via a network of proxies, has conducted a series of online influence campaigns to muddy the narrative waters in an attempt to convince users in English-speaking countries to pushback against the prevailing factual narratives associated with Xinjiang.
In May 2021, the Oxford Internet Institute published a research study called China’s Public Diplomacy Operations, which found that state-run media entities and Chinese diplomats are frequently retweeted by “an army of fake accounts” on Twitter. The OII study found that, “more than one in ten of the retweets of PRC diplomats between June 2020 and January 2021 were from an account later suspended by Twitter,” for platform violations.
In order to demonstrate how these influence campaigns work and the tactics they employ, we analyzed recent conversations using two hashtags that saw spurts of usage throughout early 2021 promoting the Chinese line on the treatment of Uyghurs in Xinjiang: #stopxinjiangrumors and #supportxinjiangcotton. The hashtag #stopxinjiangrumors was deployed to promote the idea that Uyghurs in Xinjiang are happy and prosperous. The hashtag #supportxinjiangcotton is geared toward combating recent efforts to boycott goods using cotton from the region on the grounds that it is farmed using forced Uyghur labor.
In late January, a number of accounts began pushing the hashtag #stopxinjiangrumors, specifically targeting then Secretary of State Mike Pompeo, @-mentioning and hashtagging his name. The campaign started less than two weeks after the U.S. State Department released a statement referring to China’s actions against Uyghur people in Xinjiang as “genocide.” Tweets using the hashtag, which continued into March, generally sought to disparage Pompeo as unreliable and present life as a Uyghur Muslim in Xinjiang—told often in the first person with no real indication of authentic identity—as highly enjoyable and entirely free.
The hashtag was revived in June and July to push additional propaganda videos.
In March, another group of accounts launched a new hashtag with less longevity but higher spurts of engagement: #supportxinjiangcotton. The hashtag campaign suddenly appeared in the immediate aftermath of Western countries implementing sanctions on China over the role of forced Uyghur labor in Xinjiang cotton production, thereby reinserting major apparel brands like H&M and Nike into the debate.
Both hashtag campaigns began under suspicious circumstances. The #stopxinjiangrumors campaign was initially pushed by an account that exclusively promotes Xin Jiang tourism attractions and the supposed wellbeing of Uyghur people there, while #supportxinjiangcotton was first used by an editor at a Chinese state-run media organization.
As we noted in July, one common signal of possible social media coordination campaigns is an abnormal distribution of account creation dates, suggesting a dedicated effort to create a concentration of ‘sock puppets’ for the sole purpose of promoting a particular narrative. In the case of #stopxinjiangrumors, 17% of the 84 accounts examined were created in a four-day period in late January, 2020. Another 7% were made on March 18, 2020 alone. While we did not observe the same pattern among accounts sharing #supportxinjiangcotton, analysis of historical posting behavior from those accounts exhibited suspicious hyperfocus on a small set of issues: Xinjiang province, anti-vaccine propaganda (#notomandatoryvaccines, #abolishbigpharma) and anti-American sentiment.
Directing Traffic to Chinese-State Media
Network mapping can show the strength of relationships between accounts, and also the degree to which those accounts are promoting others and attempting to drive traffic their way. Separate networks maps of the accounts promoting these two separate hashtags showed similar patterns of promoting Chinese-State media. In the network maps below, colored nodes are accounts that have used the hashtags, while gray nodes are those that have been tagged by those accounts through mentions and retweets. The largest nodes are those that have been tagged the most, and a node’s centralized location on the map reflects its connections to various nodes across the network. We have called out a number of the accounts that appear prominently across the networks for both hashtags.
Accounts sharing #stopxinjiangrumors
Accounts sharing #supportxinjiangcotton
In examining the most highly-mentioned accounts by the promoters of each hashtag, we found the same 8 accounts among the top 20 subjects for both sets. These popular targets consist of three Chinese government officials, two state-affiliated Chinese media outlets, a critic of U.S. foreign policy, YouTube, and former U.S. president Donald Trump. The intent of such diligent efforts by these suspicious accounts is a clear attempt to get the attention of English-speaking audiences with catchy and misleading narratives, inundating them with CCP talking points that undermine American foreign policy goals.
Whether or not these accounts are centrally run, it is noteworthy that disparate groups promoting separate-but-similar hashtags consistently attempt to direct traffic to Chinese-state media and officials. In this context, it is important to acknowledge that the CCP consolidated control over all state-run media entities in 2018 when it established the Central Propaganda Department, which, at minimum, when considering the OII report’s findings, suggest some degree of correlation.
Account and network analyses on publicly available data alone cannot definitively prove that sets of accounts are centrally operated or that they serve at the behest of a particular government. The singular focus of these accounts on hyper-specific issues, coupled with incessant promotion of official Chinese sources of information, show strong signals of coordination among–and possibly between—two groups of accounts pushing pro-CCP hashtags.
Today’s online threat landscape is significantly more complex and blended, which may lead to many abusive and harmful behaviours slipping through the gaps of current moderation models. In this context, it is critical to better understand more case studies like the cotton campaign so as to develop more robust and global resiliency models that can effectively counteract these influence operations. One one hand, increased labeling of “state-run” Twitter accounts has already had an impact on user engagement with Chinese state-affiliated media, according to The China Media Project; however, these types of labels are less helpful in the whack-a-mole game of network takedowns. Understanding and defending against adversarial narratives requires analysis of both the message’s contents and context, and how they are spread through networks and across platforms.