Last month I published a piece on Raving Canuck that analyzed the Twitter follower counts of high-profile Canadian accounts. Using software from Twitter Audit (not an exact science), results showed the ratio of supposedly real to fake followers varied widely among some of Canada’s top Twitter accounts. Several journalists were intrigued enough by the piece to conduct their own audits.
A couple readers flagged a couple of other Canadian accounts that they thought looked suspicious.
Most of my previous Twitter piece was looking at the Canadian angle of a New York Times expose that revealed many prominent Twitter accounts had high fake follower counts, either purchased by the account holder or one of their affiliates from companies that provide fake followers (bots) and engagement. Although I wasn’t able to determine if some of the Canadian high-profile accounts with abnormally high fake followers counts were due to individuals having bought bots to bolster those accounts’ number of followers, the Twitter Audit data was helpful in highlighting how social media can often give false impressions. At the time the NYT expose was published, it was reported that millions of followers had disappeared overnight because Twitter had purged many of the bots connected to the reported company selling fake followers.
This report was going to include data analysis mined from the software of Social Bakers to look at prominent Canadian Twitter accounts from the time period right before and after the NYT expose came out in order to see if any of them lost a significant portion of their follower counts from Twitter’s purge, which would be a strong indicator fake followers were bought for those accounts. Unfortunately, audits on Social Bakers costs $20 a pop. Perhaps I’ll set up a crowdfunder to complete this avenue of investigation.
Beyond the Twitterverse being filled with bots, however, there are other metrics being buggered with on the platform in order to deceive users.
“I started tracking, with a high degree of confidence, that it was Devumi [company exposed by NYT story], but I wasn’t the New York Times and able to do the investigative work they did, but last March  I started reporting on a group of bots,” said John Gray in a phone interview, co-founder of Mentionmapp, a Vancouver-based company that analyzes the authenticity of Twitter engagement on Twitter, as well as engagement on other social media platforms. “I found a profile in the lead-up to the BC provincial election that was using Devumi bots to retweet and to like anti-Christy Clark tweets that this [one Twitter] author was posting.”
“With all the social platforms, if they made the barrier to joining the platform harder that’s not going to play well in terms of the number one metric that Wall Street cares about — and that’s users.”
“So to me it’s not the buying of the followers, it’s the buying of the engagement that’s actually one of the bigger issues. When I’m buying off the shelf fake profiles to amplify and to game the social metrics, to me that’s a bigger concern than just the eyeball going ‘Ooh, you’re a somebody because you’ve got 500,000 followers.'”
Gray’s methodology involves using visualization tools, machine and human intelligence to investigate the digital ecosystem. Raving Canuck had Gray look at two prominent Canadian Twitter accounts, Rebel Media founder and publisher Ezra Levant and semi-retired and former The National anchor Peter Mansbridge.
“Simplest summary – it’s like death metal versus elevator music or sitting at a lake cottage
listening to the crickets or the deafening sound of jet boats blasting by,” said Gray in his report comparing the two.
Gray’s research involved looking at data from 1,000 tweets of @petermansbridge and @ezralevant from five distinct moments in time. Gray then narrowed the analysis to accounts that engaged with one of the two accounts by tweeting at least twice. He also removed Twitter verified accounts that engaged with either Mansbridge or Levant from his analysis.
The remaining accounts analyzed in the data sets of engagement with @petermansbridge and @ezralevant were 167 and 87 profiles respectively. Gray found that 26 profiles engaging with Mansbridge’s account were “cyborgs” (bots), defined as any Twitter account tweeting an average of 50-plus tweets per day over a seven-day span. Out of the analyzed 167 profiles engaging with Mansbridge’s account, 74 were deemed real profiles, 83 as anonymous, of which 9 were grey eggs, and 10 were unclassified.
When looking at engagement with Levant’s account, Gray found 48 out of 87 profiles analyzed were classified as cyborgs. Gray found that 12 were real profiles, 66 were anonymous, 14 of those being grey-egg accounts, and 8 unclassified.
In conclusion, Gray found that 55 per cent of the profiles analyzed engaging Levant’s account were likely cyborgs. The 87 profiles engaging with Levant’s account on average tweeted 97.8 time per day in a seven-day time span. On the other hand, only 15.6 per cent of the profiles analyzed engaging Mansbridge appeared to be cyborgs. Gray also found that “12 out of the 167 profiles analyzed” engaging with Mansbridge’s account were trolling him. In Levant’s case, Gray found only one out of the 87 accounts analyzed was trolling him.
Although Gray’s Mentionmapp Analytics isn’t an exact science, it does give a window into how complex and deceiving engagement on Twitter really is. I wanted Gray to look at a range of high-profile Twitter accounts when I contacted him, but the process of analyzing a user’s engagement is a time-consuming process, so we settled on looking at Levant and Mansbridge to start. However, with the Ontario election now underway and subterfuge on social media an inevitable consequence, it might be worth Gray’s and my time to look into the engagement of the leaders of the three main parties: Doug Ford (Conservative), Andrea Horwath (NDP) and Kathleen Wynne (Liberals).