In the post-Cambridge Analytica scandal era, identifying and defusing fake news attacks on a person or a brand is becoming increasingly critical.
However, the difference between a negative message gone viral and an organized attack is not always easy to spot. Let’s take a recent Airbnb case for example, as it contains both occurrences on the very same topic.
At the end of April, a concerted attack was launched against Airbnb on political grounds. To give a bit of context, last November, following raucous requests by BDS-related movements demanding that Airbnb remove listings of apartments in Jewish settlements in Judea and Samaria to avoid being boycotted by Palestinian supporters, Airbnb removed 200 listings from its service.
Lacking expertise in the Middle East’s historical, political, tactical and legal quagmire, Airbnb’s executives spent lots of time discussing the topic with experts who routinely disagreed with each other.
In the end, the courts resolved the issue. A flurry of lawsuits was filed against Airbnb both in the US and in Israel, alleging that Airbnb’s ban was discriminatory on the basis of potential hosts and guests’ religion. Finally, early last April, Airbnb announced that it would re-list the controversial properties.
It so happened that, at the end of last April, as I started working at Communit360, I picked up Airbnb as the company to train on to familiarize myself with the software.
Puzzled by what I was seeing on the dashboard, I called my trainer who immediately suspected we were seeing a targeted, obviously organized attack campaign against Airbnb.
Hundreds of Twitter accounts had posted the exact same tweet in the course of 24 hours: Posted different tweets with the exact same content, not retweeted a tweet.
As you can see in the dashboard snapshot below, there was a cluster of 6 longtail keywords mentioned in over 600 tweets, all with the same text, appearing at the top of the neutral reference column and dwarfing any other reported mention.
Notice that the author of the original tweet took great care to avoid using any negative words or expressions that could trigger detection by Twitter’s automated filters .
This warranted additional digging to understand how and why this sudden surge took place. Was it an organic uproar, or a fake campaign?
From the dashboard, we started looking at the accounts that tweeted the “cloned tweet”, sorting them by number of followers:
The large number of accounts with zero to a one digit number of followers stood out and was clearly pointing to foul play./
Digging deeper into such accounts further confirmed that a majority of these accounts had been created or bought with the sole aim of being used in fake campaigns:
This account, for example, was created in 2011 and only used from January 2019 exclusively in support of boycott campaigns.
The campaign organizers relied on a mottled collection of genuine accounts, bought accounts, accounts created for the occasion and bots to push that cloned tweet wide and large, probably with the hope of jumpstarting a viral spread of the tweet.
Digging even deeper, we were able to identify the point of origin and the spread pattern of the tweets.
The Communit360 team contacted the relevant authorities and this specific fraudulently inflated campaign was put to an end, as can been seen below in the steep decline of neutral opinions following the end of April peak (the first one on the graph, to the left):
However, we can see two other peaks, one neutral around May 15 and one combining neutral and negative sentiment in early June.
Were these indicative of other targeted attacks or the result of genuine viral trends?
Let’s have a look at the mid-May peak first.
This time, the issue is more muddled. There is a cluster of mentions related to the BDS issue, that seems related to the most shared links.
Checking the tweets sharing those links, it appears that many are genuine. This time, the ratio of low/high follower count is lower than for the previous campaign and the fake accounts that participated in the campaign seem to be more diverse as they are protesting a wider number of topics.
Discrepancies between the account creation date and the date of the first tweet still raise doubts as to the legitimacy of some of these accounts, which actually raises another question.
The automated message that repeated hundreds of times on Twitter and on Facebook was automatically generated upon signing the petition. As a number of the accounts displaying this tweet are fake, what does it tell us about the signers of the petition? How many of the reported 20 000 people who signed the petition are real people and how many are fake? But that is another topic altogether….
In any case, the proportion of fake Twitter accounts, though far from nil, was far from as blatant as it was at the end of April.
Looking at the third peak is actually far more interesting from a trend identification perspective as it is a genuine viral trend following an Airbnb guest tweeting a video of a racial slur uttered by his host.
This was retweeted immediately by accounts with large followership and then spread like wildfire in the black Twitter community.
For brands, having the ability to spot such trends in real time provides the opportunity to evaluate their potential impact at a glance.
Zooming in on the 5 days corresponding to the peak in the timeline, we can see that the “topic by sentiment” section shows a clear dominance of tweets related to the racist occurrence.
In the “most shared link” section for the same period, all but two of the most shared links are related to that incident. This means that topic hit such a nerve that people made the effort to create a new link or video rather than simply retweeting the original post.
This topic certainly dominated the Airbnb related Twitter stream at the beginning of June.
While Airbnb cannot completely avoid the occasional unsavory host and has a policy in place to ban such hosts from using the platform in the future, this does not stop users from spreading their outrage in situations like these, especially when the report is supported by a video.
Interestingly, during the exact same period, another racist incident caused by an Airbnb host took place and was immediately addressed by @AirbnbHelp. It thus appears that Airbnb is indeed sensitive to the issue, and wishes to guarantee that guests will be treated with respect regardless of their race, gender or creed. It might very well be that they simply failed to notice the growing outrage around the other racist event taking place around the same time.
For brands with millions of customers, it is virtually impossible today to monitor all of the trends and narratives related to their brand and to properly spot and understand the origin of sudden trends pertaining to them, without relying on AI and advanced machine learning tools such as Communit360.
Only with such tools can a decision be made in full appreciation of the extent of the issue at hand, while optimally prioritizing and strategizing the proper corrective actions.