Twitter and social media are center stage during the 2016 presidential election. Democratic nominee Hillary Clinton has a large and professional social media team crafting content from campaign headquarters in Brooklyn. Donald Trump's team would like the world to believe the populist candidate rode a groundswell of Twitter-powered grassroots support to the party's nomination. Yet at the center of the social media success of both Trump and Clinton may lurk a massive botnet of fake followers.
"I became curious about presidential candidate botnets during the primary season," said political scientist William P. Stodden. Stodden earned his PhD in political science from Southern Illinois University in 2012 and currently teaches comparative politics and American government. "Trump and Clinton weren't just popular on Twitter, they were dominant. And no matter how fast Clinton's Twitter account grew, Trump's always grew a little bit faster."
READ: Donald Trump explains the thinking behind his approach to Twitter (CBS News)
Vlad Shevtsov is a financial industry metadata specialist based in Novosibirsk, Russia. "We are a small team dedicated to new data processing models," the data scientist explained during one of several Skype calls. "We have applied our analytical algorithms to the data of social networks. Our basic set of algorithms can be summed up as analysis of the context of [Twitter account] metadata. We compare hundreds of parameters to create a complete, consistent picture of [an account] object."
In late 2015 Shevtsov and his colleagues began logging and studying the metadata of presidential candidate Twitter followers. After logging Twitter data for several months, his team encountered what appeared to be a large, active network of junk accounts.
To map the shape and scale of the zombie account network, his team ran an experiment: they set up a test account and bought followers.
Twitter and other social media services have waged a long battle against web services that sell followers. Nonetheless, purchasing followers remains trivial and affordable. Some sites offer followers for as low as $5. Spending $50 per month could supply an account with thousands of zombie followers over an extended period of time.
Shevtov explained what his team learned from the experiment:
Three days after purchase, the followers number skyrocketed to the 6500 mark. According to STAQ methodology Socialpuncher.com, 94% are non-active primitive bots-followers. The facts about different sorts of bots can be found [here]... For every bot, we downloaded the set of its friends and then built the bot's social graph.
We found different Twitter accounts, many sites and a lot of gigs on fiverr.com where anyone can buy bots. If you are not close to industry and you want to grow your followers count, you probably will get it from this botnet.
We estimate this botnet in 15,000,000 Twitter accounts. This is not a single botnet, it is the complex network with several centers that receive and distribute orders. This is the largest non-dedicated botnet in Twitter. In addition, we found 3,000,000 Twitter bot accounts from other botnets.
"When I saw [Shevtsov's] data," Stodden said, "I felt a lot more confident that zombie followers might be puffing up the accounts."
We established 12 "bot-factors" and examined Twitter accounts created exclusively in 2015 following @HillaryClinton and @realDonaldTrump. A bot-factor is the percentage likelihood that a particular account is a zombie and takes into account metadata and profile information like account description, location, number of Tweets, follower-to-following ratio, and other factors.
Here's what we found:
- Between 44 and 53 percent of Twitter accounts registered in 2015 following Trump fit a bot-factor trait.
- Between 48 and 57 percent of Twitter accounts registered in 2015 following Clinton fit a bot-factor trait.
- Over 100,000 accounts following @realDonaldTrump have no description.
- 97,000 of Clinton's followers have no description.
- 38,000 accounts following @realDonaldTrump have never posted a Tweet.
- 35,000 @HillaryClinton followers have never Tweeted.
- More than 76,000 Trump followers have no followers.
- 50,000 Clinton followers have no followers.
- Accounts with no Tweets, no followers, no description, and no location are almost certainly zombies—39,000 of these accounts follow Trump and 22,000 follow Clinton.
Here's the full chart for more details:
Through the primary, Trump and Clinton grew their Twitter following at a much faster rate than other candidates. Trump added on average more than 17,000 new followers per day, and Clinton added on average 9,400 new followers. Sanders added 6,800 new followers per day, and the other candidates added on average fewer than 3,500 per day.
"These [charts] are graphical representations of the raw data," Stodden said. "It's easy to look at the spike in Trump's followers, as many in the media have, and assume the spikes are the result of real people following the candidate."
Stodden cross-referenced his social media log with Shevtsov's, along with an additional set provided by Enga.ge's Election Scorecard. "Our sets may suggest ... that many of Trump and Clinton's followers are not actual people," he said. "A large percentage are likely zombies."
The only way to force fake followers to unfollow you is to block them. When you block someone on Twitter, they're no longer able to see your Tweets or follow you (but what do they care—they're bots). To block someone, open up your follower list, click the gear next to their name, and click Block. You can see a list of users you've blocked by going to Settings > Blocked accounts.
TechRepublic's data summary is neutral. We are not implying directly or indirectly that candidates, celebrities, or companies inflated their accounts or engaged in illegal or unethical behavior. We are curious about the state of the Twitter ecosystem and the relationship between technology, politics, business, and culture.
READ: Data analytics goes mainstream (Tech Pro Research)
Through the course of the 2016 campaign in a series of stories and interviews TechRepublic has documented the growing influence of social media and big data in campaigns and elections. Our goal is to discover and share unique insights with startups and SMBs about how campaigns use tech innovation to gain competitive advantage.
If you're a botnet-buyer, data scientist, social media professional, or inquisitive TechRepublic reader we'd love your ideas about how to inspect campaign social media data. Please leave a comment below or ping us on Twitter @TechRepublic.
- Election Tech Fakers: Are candidates artificially inflating their Twitter accounts? (TechRepublic)
- How big data pioneers use open source technology to win elections (TechRepublic)
- Trump and Clinton Twitter followers are similar according to new research (ZDNet)
- L2 Profile: Big data proves that America is a purple country (TechRepublic)
- What Hillary Clinton's technology policy agenda means for business (TechRepublic)
- NPG VAN Profile: How big data pioneers use open source technology to win elections (TechRepublic)
- Campaign 2016: Our technology-enabled Bizarro World election (ZDNet)
- Harper Reed: Leadership is more powerful than technology (TechRepublic)
Dan Patterson has nothing to disclose. He does not hold investments in the technology companies he covers.
Dan is a Senior Writer for TechRepublic. He covers cybersecurity and the intersection of technology, politics and government.