Please enable javascript to access the full functionality of this site

#StopXinjiang Rumors

PB54 StopXinjiangRumours - banner
Dark

#StopXinjiangRumors

The CCP’s decentralised disinformation campaign

Introduction

This report analyses two Chinese state-linked networks seeking to influence discourse about Xinjiang across platforms including Twitter and YouTube. This activity targeted the Chinese-speaking diaspora as well as international audiences, sharing content in a variety of languages.

Both networks attempted to shape international perceptions about Xinjiang, among other themes. Despite evidence to the contrary, the Chinese Communist Party (CCP) denies committing human rights abuses in the region and has mounted multifaceted and multiplatform information campaigns to deny accusations of forced labour, mass detention, surveillance, sterilisation, cultural erasure and alleged genocide in the region. Those efforts have included using Western social media platforms to both push back against and undermine media reports, research and Uyghurs’ testimony about Xinjiang, as well as to promote alternative narratives.

In the datasets we examined, inauthentic and potentially automated accounts using a variety of image and video content shared content aimed at rebutting the evidence of human rights violations against the Uyghur population. Likewise, content was shared using fake Uyghur accounts and other shell accounts promoting video ‘testimonials’ from Uyghurs talking about their happy lives in China.

Our analysis includes two datasets removed by Twitter:

  • Dataset 1: ‘Xinjiang Online’ (CNHU) consisted of 2,046 accounts and 31,269 tweets.
  • Dataset 2: ‘Changyu Culture’ (CNCC) consisted of 112 accounts and 35,924 tweets.

The networks showed indications of being linked by theme and tactics; however, neither achieved significant organic engagement on Twitter overall—although there was notable interaction with the accounts of CCP diplomats. There were signs of old accounts being repurposed, whether purchased or stolen, and little attempt to craft authentic personas.

Twitter has attributed both datasets to the Chinese government, the latter dataset is specifically linked to a company called Changyu Culture, which is connected to the Xinjiang provincial government. This attribution was uncovered by ASPI ICPC in the report Strange bedfellows on Xinjiang: the CCP, fringe media and US social media platforms.

Key takeaways

Different strands of CCP online and offline information operations now interweave to create an increasingly coordinated propaganda ecosystem made up of CCP officials, state and regional media assets, outsourced influence-for-hire operators, social media influencers and covert information operations.

  • The involvement of the CCP’s regional government in Xinjiang in international-facing disinformation suggests that internal party incentive structures are driving devolved strands of information operations activity.
  • The CCP deploys online disinformation campaigns to distract from international criticisms of its policies and to attempt to reframe concepts such as human rights. It aligns the timing of those campaigns to take advantage of moments of strategic opportunity in the information domain.

Notable features of these datasets include:

  • Flooding the zone: While the networks didn’t attract significant organic engagement, the volume of material shared could potentially aim to ‘bury’ critical content on platforms such as YouTube.
  • Multiple languages: There was use of English and other non-Chinese languages to target audiences in other countries, beyond the Chinese diaspora.
  • Promotion of ‘testimonials’ from Uyghurs: Both datasets, but particularly CNCC, shared video of Uyghurs discussing their ‘happy’ lives in Xinjiang and rebutting allegations of human rights abuses. Some of those videos have been linked to a production company connected to the Xinjiang provincial government.
  • Promotion of Western social media influencer content: The CNHU network retweeted and shared content from social media influencers that favoured CCP narratives on Xinjiang, including interviews between influencers and state media journalists.
  • Interaction between network accounts and the accounts of CCP officials: While the networks didn’t attract much organic engagement overall, there were some notable interactions with diplomats and state officials. For example, 48% of all retweets by the CNHU network were of CCP state media and diplomatic accounts.
  • Cross-platform activity: Both networks shared video from YouTube and Douyin (the Chinese mainland version of TikTok), including tourism content about Xinjiang, as well as links to state media articles.
  • Self-referential content creation: The networks promoted state media articles, tweets and other content featuring material created as part of influence operations, including Uyghur ‘testimonial’ videos. Similarly, tweets and content featuring foreign journalists and officials discussing Xinjiang were promoted as ‘organic’, but in some cases were likely to have been created as part of curated state-backed tours of the region.
  • Repurposed spam accounts: Accounts in the CNCC dataset tweeted about Korean television dramas as well as sharing spam and porn material before tweeting Xinjiang content.
  • Potential use of automation: Accounts in both datasets showed signs of automation, including coordinated posting activity, the use of four letter codes (in the CNHU dataset) and misused hashtag symbols (in the CNCC dataset).
  • Persistent account building: ASPI ICPC independently identified additional accounts on Twitter and YouTube that exhibited similar behaviours to those in the two datasets, suggesting that accounts continue to be built across platforms as others are suspended.

The Chinese party-state and influence campaigns

The Chinese party-state continues to experiment with approaches to shape online political discourse, particularly on those topics that have the potential to disrupt its strategic objectives. International criticism of systematic abuses of human rights in the Xinjiang region is a topic about which the CCP is acutely sensitive.

In the first half of 2020, ASPI ICPC analysis of large-scale information operations linked to the Chinese state found a shift of focus towards US domestic issues, including the Black Lives Matter movement and the death of George Floyd (predominantly targeting Chinese-language audiences). This was the first marker of a shift in tactics since Twitter’s initial attribution of on-platform information operations to the Chinese state in 2019. The party-state’s online information operations were moving on from predominantly internal concerns and transitioning to assert the perception of moral equivalence between the CCP’s domestic policies in Xinjiang and human rights issues in democratic states, particularly the US. We see that effort to reframe international debate about human rights continuing in these most recent datasets. This shift also highlighted that CCP information operations deployed on US social media platforms could be increasingly entrepreneurial and agile in shifting focus to take advantage of strategic opportunities in the information domain.

The previous datasets that Twitter has released publicly through its information operations archive focused on a range of topics of broad interest to the CCP: the Hong Kong protests; the Taiwanese presidential election; the party-state’s Covid-19 recovery and vaccine diplomacy; and exiled Chinese businessman Guo Wengui and his relationship with former Trump White House chief strategist Steve Bannon. The datasets that we examine in this report are more specifically focused on the situation in Xinjiang and on attempts to showcase health and economic benefits of CCP policies to the Uyghur population and other minority groups in the region while overlooking and denying evidence of mass abuse. In both datasets, the emblematic #StopXinjiangRumors hashtag features prominently.

Traits in the data suggest that this operation may have been run at a more local level, including:

  • the amplification of regional news media, as well as Chinese state media outlets
  • the involvement of the Xinjiang-based company Changyu Culture and its relationship with the provincial government, which ASPI previously identified in Strange bedfellows on Xinjiang: the CCP, fringe media and US social media platforms by linking social media channels to the company, and the company to a Xinjiang regional government contract
  • an ongoing attempt to communicate through the appropriation of Uyghur voices
  • the use of ready-made porn and Korean soap opera fan account networks on Twitter that were likely to have been compromised, purchased or otherwise acquired, and then repurposed.

The CCP is a complex system, and directives from its elite set the direction for the party organs and underlings to follow. Propaganda serves to mobilise and steer elements within the party structure, as well as to calibrate the tone of domestic and international messaging. The party’s own incentive structures may be a factor that helps us understand the potential regional origins of the propaganda effort that we analyse in this report, and have identified previously. The China Media Project notes, for example, that local party officials are assessed on the basis of their contribution to this international communication work. It’s a contribution to building Beijing’s ‘discourse power’ as well as showing obedience to Xi Jinping’s directions.

The data displays features of the online ecosystem that the party has been building to expand its international influence. The networks that we analysed engaged consistently with Chinese state media as well as with a number of stalwart pro-CCP influencers. One strand of activity within the data continues attempts to discredit the BBC that ASPI and Recorded Future have previously reported on, but the real focus of this campaign is an effort to reframe political discourse about the concept of human rights in Xinjiang.

The CNHU dataset, in particular, offers a series of rebuttals to international critiques of CCP policy in Xinjiang. As we’ve noted, the network was active on issues related to health, such as life expectancy and population growth. CCP policies in the region are framed as counterterrorism responses as a way of attempting to legitimise actions, while negative information and testimonies of abuse are simply denied or not reported. The accounts also seek to promote benefits from CCP policies in Xinjiang, such as offering education and vocational training. The BBC and former US Secretary of State Mike Pompeo—the former having published reports about human rights abuses in the region, and the latter having criticised the party’s policies in the region—feature in the data in negative terms. This external focus on the BBC and Pompeo serves to reframe online discussion of Xinjiang and distract from the evidence of systematic abuse. For the CCP, both entities are sources of external threat, against which the party must mobilise.

Methodology

This analysis uses a quantitative analysis of Twitter data as well as qualitative analysis of tweet content.

In addition, it examines independently identified accounts and content on Twitter, YouTube and Douyin, among other platforms, that appear likely to be related to the network.

Both datasets include video media. That content was processed using SightGraph from AddAxis. SightGraph is a suite of artificial-intelligence and machine-learning capabilities for analysing inauthentic networks that disseminate disinformation. For this project, we used SightGraph to extract and autotranslate multilingual transcripts from video content. This facilitated extended phases of machine-learning-driven analysis to draw out ranked, meaningful linguistic data.

Likewise, images were processed using Yale Digital Humanities Laboratory’s PixPlot. PixPlot visualises a large image collection within an interactive WebGL scene. Each image was processed with an Inception convolutional neural network, trained on ImageNet 2012, and projected into a two-dimensional manifold with the UMAP algorithm such that similar images appear proximate to one another.

The combination of image and video analysis provided an overview of the narrative themes emerging from the media content related to the two Twitter datasets.

Twitter has identified the two datasets for quantitative analysis as being interlinked and associated via a combination of technical and behavioural signals. ICPC doesn’t have direct access to that non-public technical data. Twitter hasn’t released the methodology by which this dataset was selected, and the dataset may not represent a complete picture of Chinese state-linked information operations on Twitter.

The Twitter takedown data

This report analyses the content summarised in Table 1.

Table 1: Twitter dataset summaries

In both datasets, most of the tweeting activity seeking to deny human rights abuses in Xinjiang appears to have started around 2020. In the CNHU dataset, accounts appear to have been created for the purpose of disseminating Xinjiang-related material and began tweeting in April 2019 before ramping up activity in January 2021. That spike in activity aligns with the coordinated targeting of efforts to discredit the BBC that ASPI has previously identified. While some accounts in the CNCC dataset may have originally had a commercial utility, they were probably repurposed some time before 19 June 2020 (the date of the first tweet mentioning Xinjiang and Uyghurs in the dataset) and shifted to posting Xinjiang-related content. Former Secretary of State Mike Pompeo gave his attention-grabbing anti-CCP speech in July 2020, and criticism of him features significantly in both datasets.

Previous ASPI analysis identified Twitter spambot network activity in December 2019 to amplify articles published by the CCP’s People’s Daily tabloid, the Global Times (figures 1 and 2). The articles that were boosted denied the repression of Uyghurs in Xinjiang and attacked the credibility of individuals such as Mike Pompeo and media organisations such as the New York Times. It isn’t clear whether that network was connected to the CNHU and CNCC datasets, but similar behaviours were identified.

Figure 1: Tweets per month, coloured by tweet language, in CNHU dataset

Figure 2: Tweets per month, coloured by tweet language, in CNCC dataset[fig2]

An overview of the tweet text in both datasets shows that topics such as ‘Xinjiang’, ‘BBC’, ‘Pompeo’ and ‘Uyghur’ were common to both campaigns (Figure 3). While there were some tweets mentioning ‘Hong Kong’, specifically about the Covid-19 response in that region, this report focuses on content targeting Xinjiang-related issues.

Figure 3: Topic summary of tweet text posted between December 2019 and May 2021

In early 2021, the #StopXinjiangRumors hashtag was boosted by both networks. Accounts in the CNHU dataset were the first to use the hashtag, and many accounts potentially mistakenly used double hashtags (‘##StopXinjiangRumors’). Accounts in the CNCC dataset that were batch created in February 2021 appear to have posted tweets using the hashtag and tagged ‘Pompeo’ following the tweets posted by accounts in the CNHU dataset. The use of the hashtags may be coincidental, but the similarity of timing and narratives suggests some degree of coordination. #StopXinjiangRumors continues to be a hashtag on Twitter (as well as YouTube and Facebook).

The rest of this report presents the key insights from the two datasets in detail.
 

Dataset 1: CNHU

Dataset 1: CNHU - Key points

  • Nearly one in every two tweets (41%) contained either an image or a video. There were in total 12,400 images and 466 videos in the CNHU dataset.
  • This video and image content was aimed broadly at pushing back against allegations of human rights abuses in Xinjiang, particularly by presenting video footage of ‘happy’ Uyghurs participating in vocational training in Xinjiang, as well as screenshots of state media and government events promoting this content.
  • The network promoted phrases commonly used in CCP propaganda about Xinjiang, such as ‘Xinjiang is a wonderful land’ (新疆是个好地方)—the eighth most retweeted hashtag in the CNHU dataset.
  • In total, 48% (1,308) of all retweets by the network were of CCP state media and diplomatic accounts. The Global Times News account was the most retweeted (287), followed by the account of Ministry of Foreign Affairs (MOFA) spokesperson Hua Chunying (华春莹) (108).
  • While the network shared links to state media, YouTube and Facebook, many videos shared in the CNHU dataset appeared to have originated from Douyin.
  • The network worked to promote state media. Of all the tweets, 35% had links to external websites—mostly to Chinese state media outlets such as the China Daily, the China Global Television Network (CGTN) and the Global Times.
  • The network showed potential indicators of automation, including coordinated posting, the appearance of randomised four-letter digit codes in some tweets, and watermarked images.
  • The network tweeted and shared content in a variety of languages, including using Arabic and French hashtags, suggesting that it was targeting a broad audience.

Dataset 2: CNCC

Dataset 2: CNCC - Key points

  • The CNCC dataset contained a considerable amount of repurposed spam and porn accounts, as well as content linked to Korean music and television.
  • While there was a small amount of content about Hong Kong and other issues, most of the non-spam content related to Xinjiang. Much of that content sought to present ‘testimonials’ from Uyghurs talking about their happy lives in China.
  • Some of this content may be linked to a company called Changyu Culture, which is connected to the Xinjiang provincial government and was funded to create videos depicting Uyghurs as supportive of the Chinese Government’s policies in Xinjiang.
  • The network had a particular focus on former US Secretary of State Mike Pompeo: @蓬佩奥 or @‘Pompeo’ appears 438 times in the dataset. Likewise, video content shared by the network referenced Pompeo 386 times.

Download Report & Dataset Analysis

Readers are encouraged to download the report to access the full dataset analysis. 

ADF

Australian Defence Force

ACSC

Australian Cyber Security Centre

IEC

the International Electrotechnical Commission

IEEE

Institute of Electrical and Electronics Engineers

IoT

Internet of Things

IoTAA

Internet of Things Alliance Australia

ISO

International Organisation for Standardization

USB

universal serial bus

IIOT

Industrial Internet of Things

ASD

Australian Signals Directorate

CCP

Chinese Communist Party

MERICS

Mercator Institute for China Studies

PRC

Peoples Republic of China

VPN

virtual private network

AI

Artificial Intelligence

SCS

Social Credit System

BRI

One Belt, One Road initiative

CETC

China Electronics Technology Group Corporation

NGO

nongovernment organisation

RFID

radio-frequency identification

CFIUS

Committee on Foreign Investment in the US

SVAIL

Silicon Valley Artificial Intelligence Laboratory

UTS

University of Technology Sydney

ATO

Australian Taxation Office

COAG

Council of Australian Governments

DHS

Department of Human Services

DTA

Digital Transformation Agency

FIS

Face Identification Service

FVS

Face Verification Service

TDIF

Trusted Digital Identity Framework

NUDT

National University of Defense Technology

PLAIEU

PLA Information Engineering University

RFEU

Rocket Force Engineering University

STEM

science, technology, engineering and mathematics

UNSW

University of New South Wales

ZISTI

Zhengzhou Information Science and Technology Institute

AFP

Australian Federal Police

ACIC

Australian Criminal Intelligence Commission

NATO

North Atlantic Treaty Organisation

A4P

Action for Peacekeeping

ASEAN

Association of Southeast Asian Nations

C-34

Special Committee on Peacekeeping Operations

CTOAP

Peacekeeping Training Centre (Timor-Leste)

F-FDTL

Timor-Leste Defence Force

MFO

Multinational Force and Observers

MINUSCA

UN Multidimensional Integrated Stabilization Mission in the Central African Republic

MINUSMA

UN Multidimensional Integrated Stabilization Mission in Mali

MONUSCO

UN Stabilization Mission in the Democratic Republic of the Congo

PNGDF

Papua New Guinea Defence Force

PNTL

National Police of Timor-Leste

RAMSI

Regional Assistance Mission to Solomon Islands

RFMF

Republic of Fiji Military Forces

RPNGC

Royal Papua New Guinea Constabulary

RSIPF

Royal Solomon Islands Police Force

UNAMI

UN Assistance Mission for Iraq

UNAMID

UN–African Union Mission in Darfur

UNAMIR

UN Assistance Mission for Rwanda

UNAVEM

UN Angola Verification Mission

UNDOF

UN Disengagement Observer Force

UNIFIL

UN Interim Force in Lebanon

UNIKOM

UN Iraq–Kuwait Observation Mission

UNIOGBIS

UN Integrated Peacebuilding Office for Guinea-Bissau

UNISFA

UN Interim Security Force for Abyei

UNOSOM

UN Operation in Somalia

UNMHA

UN Mission to Support the Hodeidah Agreement

UNMIBH

UN Mission in Bosnia and Herzegovina

UNMIK

UN Interim Administration Mission in Kosovo

UNMIL

UN Mission in Liberia

UNMIS

UN Mission in Sudan

UNMISET

UN Mission of Support to East Timor

UNMISS

UN Mission in South Sudan

UNMIT

UN Integrated Mission in East Timor

UNOTIL

UN Office in East Timor

UNSMIS

UN Supervision Mission in Syria

UNTAC

UN Transitional Authority in Cambodia

UNTAES

UN Transitional Administration for Eastern Slavonia, Baranja and Western Sirmium

UNTAET

UN Transitional Administration in East Timor

UNTSO

UN Truce Supervision Organization