During the height of Melbourne’s lockdown, as the city struggled to contain the virus, a separate battle was being fought online.
On Twitter, it centred around rival hashtags backing or attacking Victorian Premier Daniel Andrews’ handling of the coronavirus outbreak.
On one side: #DictatorDan and #DanLiedPeopleDied.
On the other: #IStandWithDan.
Popular opinion mattered more than ever – controlling the virus required Melburnians to follow the lockdown rules – and a trending hashtag may have appeared to reflect widespread views.
In reality, however, those hashtags were little more representative than billboard ads purchased on the sides of roads.
New research shows they were driven by a small number of fringe, hyper-partisan accounts – many of them anonymous “sock puppet” accounts created specifically to support either side, but posting as an independent third party.
In a paper published in Media International Australia, researchers found these accounts essentially “gamed” Twitter to get their hashtags trending and give the impression of a groundswell of public opinion fiercely supportive or critical of the Premier.
“People have figured out how to game this system, so with a very small number of very active accounts, they can promote a specific hashtag,” said lead author Timothy Graham, a senior lecturer in digital media at Queensland University of Technology (QUT).
A “vicious feedback loop” saw television and newspaper commentators pointing to trending hashtags as evidence of what ordinary Melburnians were thinking.
This then encouraged and appeared to vindicate the accounts pushing the hashtags.
At its height, the apparent popular revolt threatened to undermine the lockdown measures themselves.
“History could have played out differently for Melbourne,” Dr Graham said.
“This is not to diminish the right for people to reflect their views. It’s not that it’s all fabricated – there were legitimate questions about the Victorian government’s handling of the outbreak.
Accounts set up to tweet a specific issue
Pro-Andrews hashtags showed patterns of “inauthentic behaviour”, but the anti-Andrews hashtags were considerably worse, Dr Graham said.
More than half of the top 50 tweeters for #DictatorDan and #DanLiedPeopleDied were identified as sockpuppets, compared to 34 per cent for #IStandWithDan.
The pro-Andrews hashtag, however, had an equally high concentration of tweets sent from a small number of very active accounts.
“The top one per cent of contributors sent one-third of the total number of tweets for #IStandWithDan,” Dr Graham said.
Over a third of all #DanLiedPeopleDied retweets were retweets of only 10 unique accounts.
The anti-Andrews hashtags also had double the number of recently created accounts to #IStandWithDan, suggesting they had been set up to tweet about that specific issue.
Dr Graham said it was not clear if any central organisation or individual had organised the accounts promoting the hashtags.
But he noted that the narrative of an authoritarian and untruthful premier aligned with the “co-ordinated anti-Andrews campaign” in News Corp papers and on Sky News.
“Based on what we can see, this strong narrative alignment is not happening by chance,” he said.
“[But] that’s not enough to say it’s driven by anyone in particular.”
From fringe to front page
The example of #DanLiedPeopleDied shows how a few active accounts can make a fringe view trend on Twitter and, ultimately, make front-page news.
The hashtag had been circulating at low volume since March 2020 among a group of hyper-partisan accounts.
Then on the morning of August 12, one of these accounts spearheaded an orchestrated campaign to push the hashtag onto the trending topics list, Dr Graham said.
“And then Avi Yemini saw it the next morning.”
Extreme far-right political figure Avi Yemini (who was banned from Facebook and attracted controversy because of his extremist views and criminal history) tweeted the hashtag seven times to his 128,000 followers that day.
“That morning it exploded,” Dr Graham said.
Of the 6078 tweets posted on August 12, almost 1000 were retweets of Yemini.
Over the following weeks and months, Yemini attracted one in 10 of all retweets for this hashtag.
An inquiry into the failures of Victoria’s hotel quarantine that sparked the state’s second wave found no evidence that Premier Andrews had lied, though it did highlight serious failures in program set-up and management.
Daniel Angus, a co-author of the paper, said the fact a small number of accounts could push an issue to trend showed Twitter’s mechanics were “broken”.
“It’s almost a game played by people to get certain hashtags trending on a certain day,” he said.
“An issue can go from a few fringe accounts to front-page splashes – a few individuals can steer the public discourse.”
Little evidence of bot activity
Despite popular speculation that the hashtags were being pushed by networks of automated bot accounts, the QUT research found little evidence of this.
The team ran a “state-of-the-art” bot detection tool over the top 1000 accounts for each of the three hashtags, Dr Graham said.
“We basically found very few bot accounts on either side.”
For #DictatorDan, only 25 of the top 1000 accounts were bots, and the other hashtags had an even lower figure.
Dr Graham said the problem went deeper than the proliferation of bot accounts to the nature of online engagement itself.
“This stuff is good for business. When these types of emotionally charged hyper-partisan discussions emerge, people eat it up.”
In the US, disinformation scholars called on Twitter to hide its trending feature in the run-up to the November 3 US election, arguing that trends could serve as a vehicle for disinformation.
“I’d say about four times out of five, hashtags that appear on there are genuinely of interest – it’s what people are talking about,” Dr Graham said.
“It’s this other 10 to 20 per cent that’s problematic.”