Taiwan 2020: Monitoring a democracy under attack

On January 11, 2020, the incumbent Taiwanese president, Tsai Ing-Wen, surged back into power with a massive majority.

Known for her liberal democratic stance, Tsai has long been a bulwark against increasing pressure from China for a closer relationship with Taiwan.

China views Taiwan as part of its territory, under the ‘One China’ principle. It has a long-term vision of taking the island back. Tsai stands against this.   

The Chinese authorities were not pleased with Tsai winning. After election day, Chinese government officials cast doubt on the integrity of Tsai’s win, accusing her of dishonest practices.

The main opposition candidate, Han Kuo-Yu, had sought closer ties with Beijing. Had he won, there would have been major changes for China-Taiwan relations.

The stakes were very high.

In the weeks leading up to the election, intense online campaigning activities created a fertile environment for interference and manipulation.

Working alongside a team of OSINT and disinformation experts, I monitored online discussions around the election.  

In this post, I’ll discuss some of the main disinformation narratives that we discovered around the two main candidates.

Before exploring the data, we needed to become familiar with the wider context of the election, including basic political dynamics, key players, who sympathised with who (e.g. Han Kuo-Yu = pro-China), and any significant stories. 

I spent a chunk of time at the beginning reading about Taiwan’s recent political history. China is a key player here, but the US, Hong Kong and Japan have also influenced Taiwan. Hence, any of them could be included in false narratives.

Key Adversarial Narratives

The political dynamics of Taiwan suggested that disinformation would probably favour Han while trying to discredit Tsai.

Attribution of info ops is never easy. But it seemed likely that China-linked operatives would produce anti-Tsai and pro-Han material.

Here are some examples of the most persistent disinformation narratives we found throughout the campaign.

#1. Tsai’s Fake Doctoral Thesis

Tsai Ing-Wen holds a PhD from the London School of Economics. Questioning the credibility of her doctorate was one of the most persistent disinformation narratives of the whole election campaign.

While other rumours came and went, the one about Tsai’s supposedly fake PhD just kept on going. We found mentions of it across Twitter, Facebook, and YouTube, plus on Taiwanese discussion forums and content farm websites.

#2. The Chinese Defector

News of Wang Liqiang, the Chinese spy who defected to Australia, created a wave of activity among pro-China Twitter accounts. It’s a good example of a true story being leveraged as part of a biased narrative.

We identified a group of Twitter accounts working in tandem to spread anti-Tsai content, using the Wang story as a tool to discredit President Tsai Ing-Wen. 22 of the accounts were single-topic, focusing only on anti-Tsai posts.

The other seven accounts combined similar anti-Tsai posts with posts about other pro-China narratives, including support for the Hong Kong police against the protestors.

#3. Han And The Crying Baby

Disinformation narratives targeted pro-China candidate Han Kuo-Yu less frequently than Tsai. This was even despite the supposed existence of a Tsai-led cyber army, which might have been expected to attack him.

In late December, a story about Han appeared in the China Times. It attempted to ‘debunk’ an earlier negative incident involving Han.

According to the story, Han attended a toddler crawling competition. While there, he had picked up a baby girl without permission from her parents. The baby started to cry. But the baby’s parents later claimed that this news was fake.

Han supporters speculated that his polling numbers would improve in the aftermath of what they claimed to be a smear attempt orchestrated by the DPP (Tsai’s party).

We found examples on social media of the story being turned around to discredit Tsai in this way. Accounts claimed that the DPP ‘cyber army’ had targeted Han with disinformation about the crying baby.

Challenges

  • Getting up to speed quickly on a new and unfamiliar political context.
  • Having to handle large volumes of content in traditional Chinese characters. 
  • Not having the same ability to quickly skim read (as would have had in English), important for spotting anomalies and patterns.
  • Having to rely on linguists to provide nuance to the content, adding an extra layer of complexity.

Analysis Methods

  • Spotting Repetition: Looking for repeated images and chunks of text in social listening search results is a very useful indicator of inauthentic behaviour. When the goal is amplification of a certain message, lots of accounts work in tandem to post it across numerous channels.
  • Identifying Suspicious Communities: We found groups of Twitter users who followed only each other and interacted with each other’s content. Most showed signs of being suspicious; such as having only a handful of followers, no profile image (or one that looked inauthentic), and no bio text.
  • High Engagement: We searched political Facebook groups for highly engaged posts about known disinformation narratives, then examined the list of who shared them. We were looking for users who had shared the same post to many other groups (e.g. to a long list of pro-Han groups).
  • Tracking shared URLs: We identified URLs leading to suspicious political content, (e.g. from known content farms, or vehemently anti-Tsai). We then used social listening tools to find all the users who had shared those URLs. This often led to discovering communities of seemingly inauthentic users, especially useful on Twitter.
  • Reverse Image Search: Using TinEye, Yandex or Google to search for memes often led to a slew of results showing where the image had been posted around the web. We could then follow the trails to whichever platforms they led us to.

Similar Posts

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.