In Part Three, we introduced a set of political Twitter screen_names and user_descriptions that are hyperactive (100+ tweets/day), sharing known English-language propaganda from the Russian state-sponsored disinformation sources, active in the #MAGA movement and other western democratic activity online, but also noticeably present in viral campaigns like #KamalaHarrisDestroyed. The argument is not about the overwhelming numerical dominance of hostile bots and trolls flooding Twitter with fake news and propagandist messaging, though that is a certainly one sub-plot of the unfolding history. In this post, we look for signs that bots and trolls are active in other parts of the Canadian election discourse, comparable to their activity in #TrudeauMustGo.
Ideological Warfare on Twitter
As the blog series progresses, we will introduce data visualizations that document over time the role of agent provocateurs seeding and propelling many niche ideological communities. There is no singular disinformation strategy to be unveiled. Rather, it is apparent that political actors will try as many tactics as possible to see what works with specific target demographics. The core retweet and like functionalities in Twitter provide an instant measure of uptake for a particular ideological message, meme, image, link, or other networking bait.
Twitter inadvertently engineered a platform for A/B testing ideological warfare. In truth, it is A/B/C/D/E testing on toward infinity. Perhaps this was always the nature of free speech. However, as numerous scholars have noted, the architecture of affordances in a social media platform is fundamentally different than a spoken word on a street corner, a printed word, or even a paid advertisement on television.
In the later parts of this series, I will write about some policies (let’s call them disaffordances for now) that might constrain the spread of disinformation by hostile actors while preserving the fundamental right to free speech. We can probably all agree there is no way to eliminate or regulate useful idiots online. Maybe we can further agree to take the freely available automatic weapons of disinformation warfare out of their hands via regulation.
Twitter Trains in the TrudeauMustGo Data
As is often the case, when you spend time with data, some features leap off the page at you. Take the Twitter Train example, where “clout comes from [a] presence in some of the leading pro-Trump “rooms,” private spaces on conservative Twitter that allow followers to coordinate messages and then retweet each other—dramatically multiplying their impact.”
The public version of Twitter Trains was one of the first features I noticed in the TrudeauMustGo data, though it was an accident. The default listview in DiscoverText sorts messages longest to shortest. When you tag 49 other people in a tweet, or one of the 50 replies to that tweet, a lot of notifications pop in people’s Twitter accounts. This helps promote further engagement and seeds the possibility for new mutual follows among like-minded account holders. This is what a Twitter Train looks like in the TrudeauMustGo data:
Twitter Trains in Data on Other Canadian Candidates
So far, we have been primarily focused on the TrudeauMustGo campaign. Do we also see Twitter Trains in three primary candidate mention streams? Here is the top of the list for “AndrewScheer” and below are several illustrative examples:
The first three on the list:
Look closely at the third example above, which is from the Twitter user going by “just_my_opinion“, a Twitter account that has recorded almost 10,000 tweets since January 2018 and is retweeting at an extraordinary rate leading into the Canadian election. Among the 330 followers of this account are almost exclusively accounts that have political ideology as their defining bio theme. For such an ardent political communicator, one wonders why “just_my_opinion” follows only 110 accounts yet still attracts 330 ideologically coherent followers? Also, is this an authentic locution for a purported Alberta resident? “…I bet I was barely on solid food when I learnt my parents see through lies. The good old days, when a parents hand could slap both cheeks at the same time.” It sounds to me like it was written by someone using English as a second language.
The public Twitter chains (defined here as tagging 10 Twitter users in a post) are found in all four major Canadian election datasets. Note they are least prevalent in the TrudeauMustGo set.
|Archive Name||Total Tweets||Tweets With Trains||Percent Train Tweets|
Like many parts of this research, it is when you dive deeper into a single facet of the data that things you are not looking for emerge. In this case, one tweet that caught my attention is from https://twitter.com/themadsloth. She is apparently researching lists of Twitter handles as well, claiming “she” is on a “liberal watchlist!!! Check if your name is here.” BTW: “munch289” is everywhere in this research.
Having never been invited to look into any mad sloth before, I thought we should plug the account into botometer and see if anything pops up in the accounts followers. With about 0.0% surprise, I learned (“learnt”?) that this mad sloth with >159,000 tweets in fact has lots of suspected bot followers:
The interplay of trolls and bots in this space is undeniable. Each time we plug a suspicious account (troll or bot) into botometer, or look manually at its followers, it is the follower accounts that cause much of the concern.