Is Twitter brave enough to tackle fraudulent practices by some of its users?
Spanish journalist Jose Maria Mateos has followed up (article in Spanish) a January article in The New York Times I too mentioned a few days ago, entitled “The Follower Factory”, which detailed the practices of companies like Devumi, which boost the popularity of Twitter account holders by creating fake followers, using retweet campaigns, etc. Using Python and R, Mateos looked at several Spanish accounts — among others mine.
Python and R can reveal suspicious patterns suggesting the purchase of followers on Twitter, but cannot detect, for example, that users have artificially retweeted something to increase its visibility or viralize it). In short, the presence of these patterns does not prove that someone has used a followers service (someone else could have bought them): alterations in the graph only indicate that “somebody, somewhere, did something.”
Seen in this light, Mateo’s results are not as “sexy” as we might have hoped, and are based simply on a tool that allows one to create a simple time graph based on ordinates and the number of followers on abscissas, and in which the appearance of horizontal lines indicate that an account has experienced a sharp rise in followers: the more horizontal the line, the more direct. What’s interesting about this is that it allows me to contrast my account with what I know I did over time. In the case of my Twitter account, I publish relatively little besides referring to my daily blog in Spanish and English, occasionally passing on news or answering a question: I’m not interested in boosting my popularity, although obviously, I appreciate the fact that people read my stuff, despite being a humble academic.
One thing I do know: I have never used any tool to buy followers, not even in the interests of scientific research. Can I guarantee that somebody else hasn’t done so without my knowledge? No, but it would be very unlikely in my case, since I have never participated in any tweet payment or sponsorship scheme, which are probably quite common in other profiles: influencers, for example, who want to build a profile that simulates relevance and who are prepared to pay for those followers themselves, or people who decide to participate in a sponsored tweets campaign for a brand, which then decides to pay to increase that person’s followers to get a better return on their investment.
José María Mateos describes my graph as “quite normal”. The absence of sharp horizontal lines would suggest no manipulation, although there are some periods, two specifically, in which followers are incorporated faster than usual, specifically between 2009 and 2010 and, to a much lesser extent, between mid-2014 and 2015. I have already highlighted the first case, which coincided with the launch of Twitter in Spanish: my account was recommended by Twitter, prompting many recently created accounts at a time when the social network was growing rapidly, to follow me. At the time, I called on Twitter to eliminate the practice of recommending users. I have no idea what caused the other, more recent blip, although I have not spent much time analysing it.
Where am I going with this? In response to my entry on February 1, I received an email from Twitter’s communications director in Spain, Elena Bule, responding to my criticism that the company was not doing enough to tackle the problem of buying followers. She outlined Twitter’s efforts in this regard. The truth is that Elena’s mail came on a particularly busy day, and given that we have a pretty good relationship, I sent back a snappy reply saying that Twitter’s steps would have no impact, because buying followers is standard practice, even though the company has banned it. Replying with far greater politeness than my response merited, Elena invited me to a meeting with her security team and country manager to discuss the issue, which has yet to take place, but at which I will lay out my arguments.
My point here? That while I know what I have done or failed to do with my Twitter account, it’s still hard to explain some of the changes Python and R register, which could be down to Twitter. In other words, if Twitter wanted to isolate patterns revealing the purchase of followers, it could do so with the right algorithms. Is it possible to distinguish the patterns created by the purchase of followers from those related to a surge in popularity, such as might happen after a major news event? If, for example, somebody suddenly attracts a lot of followers and is given major media coverage, the number of people following them is likely to rise over the next few days, which might be interpreted as them having purchased followers. Having said that, more detailed analysis of events and correlations, for example, mentions on Google News, could be used. This might prove tricky to do externally, but would be easy enough from within the company, with all the data in hand.
My work analysing time series gives me optimism in this regard, and I hope my conversations with Twitter reflect this. If Twitter wants to regain its relevance, it must take serious measures against fraud: buying followers, as well as eliminating accounts that continue to do so, regardless of the short-term impact on subscribers. Whether or not the company is brave enough to grasp the bull by the horns is another matter.
Watch this space.
(En español, aquí)