Pathology of a Fake News Story

How a neutral piece of coverage on the Syrian war is reframed and distributed to misinform the public— a step by step explanation.

Fighters of the Euphrates Liberation Brigade in Syria, October 2016. From Wikimedia by Kurdishstruggle; rights CC-BY-2.0.

The accuracy of our media has been placed under constant question, with many claims being called out as false, or fake. Sometimes, this is done because someone doesn’t like a story or fact in the media; other times, the news really is fake.

This post shows a story originating in the Middle East, about Russian soldiers clearing up bombs left in Syria by Obama’s troops. The story was related using first-hand video and personal accounts, and was picked up by major outlets. However, the truth was that this story was completely false — fabricated and framed in such a way that it looked like real news. We’ll pull on threads behind this fake news, and find just one small part of what may well be a large, international network that is feeding our Western media.

I’ve been working on rumors, fake news and so on for a long time. In 2012, we proposed a multinational project on truth in online media, Pheme, which was funded with €4.3M by the European Commission. The project lasted a little over three years, finishing Spring 2017. We spent the time building definitions of fake news, models of rumors, and tools for picking up untrue claims (a tough project!). We even ran an evaluation where teams from all over the world tried to pick out fake stories from real ones, RumorEval. This was perfect timing, with 2016 being the explosive year that it was and “fake news” becoming a hot term just as our project reached its peak.

In the project, we take in huge amounts of data from the web, streaming in various social media sites, news sites, and so on. These need to be linked and grouped so that our journalists (from can readily digest content and see what claims and stories are emerging. This grouping and linking is tough, and to understand how to do it, you have to look at the data really close up. And when you do that, you start to see recurring patterns and behaviors in the tweets and stories on the web.

Let’s start with one story. This is, as it goes, a fairly neutral bit of data; it’s video footage from the destroyed Syrian city of Aleppo, of soldiers in Russian uniforms operating a piece of equipment in the streets. The caption tells us that these are Russian soldiers, using a bomb-disposal device to make the streets safer. This footage is captured and relayed by Ruptly TV, a German-based media organisation, on January 4, 2017.

In fact, it will become the first of four stages going from legitimate news to honest, but incorrectly informed, political belief.

  1. Generation: On-the-ground reporting gives new content
  2. Re-framing: Details or small changes are added to the content, making a legitimate-sounding but unsubstantiated or false claim
  3. Distribution and discussion: The re-purposed story is used in discussion and spread across the web, largely by unwitting members of the public, though seeded and spread by a core of minor outlets
  4. Fake news: News sources publish articles on the deliberately re-purposed claim

The video story is on its own fairly benign. It casts the Russians in something of a positive light, and it’s hard to see what’s wrong with people disposing of a bomb; let’s give credit where it’s due.

Social media users and, later, online news outlets, in typical fashion, pick up the story and relate it to their followers. This continues and the story disperses, but typically isn’t set into any kind of “frame” — portrayed as being the fault or success of no-one in particular. Here’s a bit of content from social media users, and an Iran-based outlet, PressTV, who distribute in Persian, English and French. Pretty neutral stuff, when it comes to frame.

We see a few other neutral framings of the story, but perhaps with weird connections; a Twitter author named Yusha Yuseef, working for a Syrian media outlet Al-Masdar News and writing under the name @MIG29_ — a famous Russian jet-fighter — on Twitter, also picks up Ruptly’s video.

By this point, we already have a few people calling the story out as fake, with corrections to match, though this centers on very technical aspects of the story. There’s debate as to whether the machine is really an Uran-6, or in fact a Croatian DOK-ING MVF-4. Here’s an example:

With rebuttal:

To be clear, there is no proof that these debates are staged at this early stage. But this sort of behaviour is a common tactic: distract readers and place some kind of legitimate debate in the discourse around a story, so that we are thinking not about the central claim, but something trivial and irrelevant. The fact that this discourse has happened lends credibility to the main claims; it gives the appearance that there are other people, experts, reading the story and fact-checking it as we go along. The evidence is, intentionally, always circumstantial; news manipulation is a product and to be good, needs to be hidden and deniable. Participants are careful to leave no proof, no smoking guns.

This is a known behavior in social media distractions; we saw something similar during the 2011 London Riots — take a look at this great visualization of social media around rumors. The science shows that you can guess how true a claim is from the talk around that claim (see for example this 2011 paper from Qazvinian et al., Rumor has it: Identifying misinformation in microblogs). By looking at online dialogue around a claim and measuring the proportion of support, deny, commentary or questioning, we can predict what the crowd thinks of the central claim — without having to analyse that claim directly.

Once a claim has been denied with evidence, people pile on board with that debunking, and the crowd’s attitude unifies — though we don’t know if that is because the crowd has access to the same evidence, or is just working by gut feeling. This doesn’t tell us anything directly about debunking, by the way; it just shows that we can work out the veracity of a claim by watching crowd behavior around it. Debunking is tougher, and often serves only to reinforce conflicting views, rather than neutralize them — a risky tactic. But that’s another article.

Note also that the claim about the story being fake, which is immediately debunked with evidence (much to the embarrassment of the author shouting “fake!”), comes from an account that seems to be Ukrainian. The Twitter username is VSSM_Ukraine and their profile picture has the blue and yellow coat of arms of the country. If you wanted to make a Twitter account that looked Ukrainian at a glance, so clear in fact that readers didn’t have to even look at the author profile to work this out, how would you make it look?

Note that the account shooting this claim down is quite unabashed about its purpose and orientation. It looks honest, and has nothing to hide — so there’s nothing to suspect here. Marcel Sardo is a “Pro-Russia media sniper”.

Who’s to say that this is a shill? Admitting your agenda and being honest about your intentions is unlikely to damage credibility. Even Harvard Business Review now covers using emotional intelligence to improve outcomes. Taking away the points of protest by admitting them openly increases trust, in all kinds of relationships; it’s hard to be angry at someone who admits their mistakes. To take another example: would you be more suspicious of someone selling their car who tells you the automobile is a “real snatch”, or the person who tells you that, “yeah, the starting price is a little high”? For me, the gut instinct is that the first one is rotten; while the second — if they’re clever — pretends to be honest, but has previously increased the price.

Anyway, the story gets picked up by outlets. Ruptly just provides raw material. Russia Today picks up the story, later, and gives more information. So does New China TV; they add an unsourced claim — that “the move is to assist sapper units in their mission of clearing mines and booby-traps left by the Islamic State during their retreat from the ancient city of Palmyra”. This fits Ruptly’s comments on the video, though from the actual footage, we can’t be completely sure.

Russia Today, an outlet that many Russians I’ve spoken to don’t really trust completely, also add this claim — though a little more nuanced. The precision lends credibility. The don’t attribute the bomb to Islamic State (IS), but just the previous uses of the robot:

It is the same robot which was used by Russian combat engineers to clear the Syrian city of Palmyra of explosives left behind by Islamic State (IS, formerly ISIS/ISIL) jihadists last year.

So far, this seems OK. A story is fleshed out, and there’s a bit of arguing around it. Shortly after the story breaks, though, we see quite a different framing. Twitter user Sarah Abdallah, an “Independent Lebanese writer, geopolitical analyst and commentator” with over 52 thousand followers, adds a very distinct frame.

The presentation of this author is great. It’s easy to match them with a well-known stereotype: they’ve clearly taken a particular approach in their profile photograph, flirting with the camera. There’s an Instagram account linked as well, which is standard for young females trading on their appearance (of course accounts like this run by men exist, but are way less common). A pretty girl saying educated things — an account geared up to attract attention. I want to draw attention to the word “independent” in the profile and how it’s supported by the author being from Lebanon; close enough to seem like a more credible expert on Middle Eastern events than, e.g., a Russian, American or German outlet, but far enough from the immediate area to be plausibly detached from it.

In the message, we see the video evidence from Ruptly, and direct technical terms about the activity (“sappers”, “Uran-6”), that support the authority and knowledge of the poster. We even have direct quotes, in “moderate rebels”, and the content doesn’t go against anything in the source’s claim. Though the framing is quite different: the bomb is the US’s fault. Then, of course, there are hashtags on Syria and Aleppo, to get visibility for the tweet.

And this bid for attention succeeds: almost a thousand retweets. Sarah, our Lebanese writer, presents two further alternate versions of the tweet, giving multiple angles on the story ready for consumption by other outlets regardless of the frame their readers expect.

In the third one, we take a quote from the Russian government to lend credibility, mixing fact and frame. This account is consistent in its actions, which means we can already use some machine learning models to pick it up; we can track rumour intensity over time, for examine, using something called Hawkes Process, and notice it as a common actor at periods of high rumour intensity.

From February, Sarah tweets about using a perhaps common narrative; attacking the MSM (“mainstream media”, in this case). This pushes reader trust of big outlets down, using a story described with video evidence by an apparently authoritative source.

Further content from this account (almost 2000 retweets) goes from anti-Obama to also anti-Hillary, a major figure in the US election. It’s yellow press stuff — big emotional hooks (“Won’t somebody think of the children?”) and a few photos that don’t disprove the tweet text, but certainly don’t prove its veracity. I’m no expert on army uniforms, so I can’t tell you if the soldiers are Russian or not.

Where does this news end up? Well, here are two interesting distributors on Twitter. The first is a pro-Trump pro-Russia account, @TeamTrumpRussia, based in St. Petersburg.

They support the story with more images of Russian soldiers doing things soldiers from every nation do — humanitarian outreach. Good stuff.

As an aside, the account has some great content (great as in, scientifically interesting). Take for example this story about Chechnya, a region in centuries-long conflict with Russia, and a denial that homosexuals are being detained en masse. The implication is clear, as it always is when media focus is given to a denial; instead of reporting nothing, it is softly suggested that the denial is significant, and we fill in the gaps ourselves (thinking that the denial must be a lie).

There’s also a common negotiating tactic here: anchoring. That is, outlining a point on a scale that is very unpalatable, to make a slightly unpalatable point (e.g. LGBT rights in Russia) look better in contrast. Like the hypothetical honest-seeming car salesman mentioned earlier, who went on to make concessions. Of course, the reader has to fill in enough blanks to give this account some plausible deniability, following recent official Kremlin policy on information warfare — or indeed the practices of the Truman administration and successors.

But Trump’s image is popular in Russia — this t-shirt I found on Red Square in February ’17 sold out in days. Whether the popularity is based on genuine fondness or not, I don’t know.

Moscow, February 2017. © Leon Derczynski. Usage: CC-BY

The next step in the chain is the dispersal of the story with added frame. Many people pick up the story and retweet it; these examples are twitterers from Italy and the Netherlands. The first doesn’t look much like a shill — just an honest follower and re-distributor of content, like any good Twitter user. Even I’m not certain that the account is paid, but it is incredibly single minded, and one wonders where the time and energy to run the account comes from.

The second is kind of cute; the quoted message is the Russia Today story attributing the bombs to IS, but the twitter user changes this to Obama, discrediting Russia Today in the process. This account is a bit similar to @Europe2424, which represents itself using a tattered EU flag and anti-EU slogans; working in Dutch and pushing out the same narrative.

Eventually, the message is picked up by blogs and sites all claiming to be pushing out the truth. These sites generally work by giving authoritative-looking news that fits the expectation of their readers and editor. Here are a few examples.

There’s no direct evidence that these sites work together, though many of them are very good at picking up precisely these stories. These are the sites that individuals at home go to when looking for “real” news, after their faith in the mainstream media has been successfully destroyed. Taking the source of an independent Lebanese journalist lends credibility — she’s almost an eyewitness, right? — and the video footage seals the deal.

And it’s feeding a frame that already exists. Sure, Breitbart have already been expressing this narrative for over a year before it hits Sarah Abdallah’s twitter. But the language and story are influential. For example, look at this video from CBS, a major US television network; this isn’t fake news, but the language and angle around the story has stuck. The words Obama’s “moderate rebels” are key to this imposed frame we see on Twitter. The fake news builds on these frames and encourages their presence in broad discourse.

Note that major Russian news outlets don’t touch the story. Sputnik News place it in an interest piece — “Dogs and Kites: How Russian Sappers Demine Daesh Booby Traps in Syria’s Palmyra”.

So, the message with this frame is spread. What’s interesting to look at next, then, is: which accounts are those constantly pushing out content from this point of view? These will be central to the distribution network.

It’s interesting to see how the discourse unfolds later. The existence of multiple explanations of events is turned on its head by some outlets to again discredit the mainstream media; e.g. this article on 21st Century Wire, as part of a “Fake News week” series that they maintain. Powerful naming; if everyone else’s news is fake, theirs must be, at least to their readers, the real deal. Here’s where the presence of bomb removal is contrasted with prior speculation in an attempt to discredit the UK’s Channel 4 outlet.

These would probably be the ‘last messages’ before government forces ‘annihilated’ them in #HolocaustAleppo. Channel 4 bought into this fully, even featuring a filmed ‘letter’ by Wa’ad which starts “Maybe this will be my last letter to you and the world…
the majority of the population of eastern Aleppo, reunited with the western part, celebrated their liberation, welcoming the Syrian army, and the Russians that followed with their sappers to clear buildings of mines and booby-traps left behind by the ‘moderate rebels’.

The topics covered by these accounts and outlets match up together well. The narratives are not directly linked, but we see the same clusters of frames in each case. Some of these outlets may have similar outlooks, but those engaging with the generated fake news, at the distribution stage, tend to be focused solely on these issues: anti-Obama, anti-Ukraine, anti-EU. The narrow and consistent single-mindedness is just too great for generic alt-right followers; we don’t see posts about personal lives, or feminism, or going to the gym and so on — none of the engagement one would expect from a casual person with alt right views. In fact, it looks precisely like the networks that are being found to dominate the modern information war — see for example this research on strange clusters of conspiracy talk, to be presented later in 2017.

Predictions about Brexit, Putin and Trump being linked together were novel but seemed plausible in June 2016, though one couldn’t be sure if this was just a depressing left-wing narrative. Now however, we see the concerted communication and flow of framing between so many outlets all supporting the same kind of narrative, and it’s hard to imagine that there is no organised connection behind the set of concepts (setting Ukraine to one side). Everywhere we look, there’s a manually-created explanation of events, using unsupported claims injected on top of neutral evidence, credible debates to take advantage of our gut feelings around truth and rigour, fitting together and backing each other up, buidling a powerful narrative believed up by many.

This is, in fact, very similar to a practice called “astroturfing”, which has been happening all around us for years. Imagine this. You’re an educated person with a bit of time, and you visit the doctor with a long-standing non-threatening complaint. They suggest a medicine you’ve tried before, or perhaps you’d like to try a new one. Prudently, you check out the suggested drug. You read a few articles about it, but they look like the typical marketing literature. You find four or five academic papers on the drug; most of the trials look good, with a benefit in the majority of cases. There’s one that’s inconclusive, but it’s in a less major journal, with a small sample size. You then go and check out a few review sites; most of the reviews are positive — the drug’s no panacea, but it’s better. A few of the reviews are very negative, but seem to be from people who are a little bit crazy, not using the drug right, have many other conditions, and so on. So you conclude, it looks legitimate; none of the stories look controlled, and everywhere you look, you see what you’d expect to see from a decent drug. Except it’s not. Everything you’ve read — all the end-user comments, all the peer-reviewed articles, are shills, put there by the pharma corporation to make their drug look good, and make it look good in a way that “smells” legitimate.

This astroturfing practice is well-studied; Sharyl Attkisson has a talk on it that’s worth checking out. The central idea is to place stories, evidence and so on in the locations that a knowledgeable, sceptical person would search, and give them the imperfections and angle that we all expect to see in genuine evidence.

The next step seems kind of clear: if you want to control the media, you can do so by astroturfing for journalists. Where will they look? What does good, genuine evidence look like for them? We don’t even have to worry about meeting the high standards of the Associated Press; all we need to do, as astroturfers, as manipulators, is give enough evidence in the right places so that it will be picked up by outlets that people read. Remember that we often prefer to read stories that fit our narratives, and rarely examine sources that we trust in a very critical way. Having a very high, rigorous standard of investigation isn’t necessarily related with motivating a large or important part of a population. If you want to present many voters in a swing state of the US, like Ohio, with a certain narrative, there are much more targeted — and cheaper—ways of doing this than getting a big authoritative story on CNN. Simply find a local outlet that many read, work out where their journalists are looking, and present angles and evidence that smell right.

Where to go next? Unraveling these networks and finding the common actors seems to be a good step. Looking out for those who are using a particular framing of a story can tell us which new accounts, that we haven’t seen before, are being used to spread fake news. This is now possible with modern technology; we have made advances in stance detection (checking how a crowd reacts) through things like RumorEval, and have good resources like the Media Frames Corpus to help us sift through the web automatically.

Even if the claim is exposed, the chances are that many people simply won’t believe the exposure; the narrative is powerful and a network like strong tree roots has developed naturally. Small relief may be found in the age of this problem; disruption through rumor was a problem even in classical times, with Pheme being, as well as the namesake of our project, the Greek goddess of fame, her wrath being manifest in scurrilous disrepute. Capturing and predicting the spread of the fake news does nothing to help combat it. Constant undermining of the mainstream media makes it hard for this to be a tool in repairing the damage constantly being done. Understanding the process and identifying the actors is a solid first step towards a solution.

The author has chosen not to show responses on this story. You can still respond by clicking the response bubble.