The purpose of the Troll Deflator is to analyze Kremlin troll messaging by thoroughly exploring individual Twitter posts that each represent some aspect of the millions of tweets produced by the Russian troll farm between roughly 2012 and 2018. (The troll farm is still active, but the data is more meaningful when it can be looked at from a historical perspective.)
The Twitter data comes from two places. First, there is Twitter’s archive of data from accounts that were suspended because they were linked to hostile activities by foreign governments. The scope of the Troll Deflator is limited to the first of Twitter’s Russia archives.
Enhancing that information is the work done by Clemson University researchers Darren Linvill and Patrick Warren. They were able to find roughly three million troll tweets before Twitter posted its first archive by using the list of troll Twitter handles released by the U.S. House Intelligence Committee and running them through Salesforce’s Social Studio. Their data is not as comprehensive as Twitter’s, but it provides four things that the Twitter archive doesn’t:
Additional tweets. Accounts and tweets that have been deleted by the user are not saved when Twitter creates an archive. Fortunately, some of this information was captured in the Clemson set.
Account information. When it creates its archives, Twitter uses a blunt tool method to protect users who may wrongly get designated as trolls. That is, the user names of any accounts with less than 5,000 followers are hashed, meaning that they are replaced with a really long alphanumeric code. Since most of the troll accounts didn’t reach that threshold for a variety of reasons, the Twitter archive data is hard to work with, especially when trying to connect troll personas to their internet activities outside of Twitter. The Clemson data doesn’t have that limitation.
Follower counts. The Twitter archives provide the number of followers at the time an account is suspended, but the Clemson data shows the number of followers the troll had at the time of the tweet.
Linvill and Warren categorized troll accounts in a really useful way. They looked at patterns of activity and messaging for each account in order to assign types to them, and then they analyzed each type as a group. Here at the Troll Deflator, we work primarily with the types that the Clemson researchers call Right Troll and Left Troll. Essentially, a Right Troll pretends to be a conservative American and seeks out a conservative American following, with the general goal of pushing the right further right. A Left Troll does the same thing but working on the other end of the political spectrum. That’s a simplification, though. The Troll Deflator is here to explore the complexities.