In Part Six, we talked about the most active day on Twitter in a collection of Tweets from the Search API for the three Twitter handles MaximeBernier, JustinTrudeau, and AndrewScheer. Here the focus is a single hour (the most active) on that single day.
In this post we introduce social network graphs for the first time to surface connections in the data.
The Busiest Hour on the Busiest Day
Below we see the distribution of Tweets over time that are the slice of election data in focus today.
Start by noting there is no “spike” in the hour that itself is the biggest spike on the most active day. What sorts of accounts were most active at 3:00 AM St. Petersberg time? What are their network properties?
TopMeta Values in the Data
We use summaries of the “TopMeta” to hunt and peck for clues.
Given the prominence of Maxime Bernier in the user_mention field, it is worthwhile noting that the CBC Poll Tracker has Bernier at 2.5% support.
The CBC projects Bernier’s party may pick up 1 seat in parliament. Yet in this data sample, his Twitter handle is far and away the most mentioned, followed at #2 by Dave Rubin, who writes Tweets attacking Antifa, the New York Times, and liberals in general, while defending Tulsi Gabbard and mocking the connection her prominence has to hostile foreign trolls and bots. At #3, we have definitely seen “MrAndyNgo” throughout the Canadian election data; the account is also pro-Trump and obsessed with Antifa. The Twitter account ThinGrayLine01 comes in at #4 garnering even more mentions than “JustinTrudeau” and propgating content like this:
What Happened 3:00-4:00 AM St. Petersburg Time?
To better understand the composition of the biggest spike in Tweets on the most active day, we loaded the tweets from this time period into a local Neo4J graph database, and analyzed them using GraphXR from Kineviz. We loaded the data in by taking all the tweets from the one-hour spike and extracted the metadata for both the user and the tweet for every item.
From this data, we further extracted and formed relationships between the tweeting user and any user mentions, any user retweets, as well as any website mentions. This formed a fairly comprehensive graph around these most active users and allowed us to explore and visualize better the connections between these actors. In this one-hour sample, there are a total of 6,591 nodes generated by 6,567 Twitter users.
Let’s zoom in and focus first on the users who made >=10 mentions during this hour.
There are just 51 users total in this group and the node size is by pagerank. Look at the central point in this graph: ThinGreyLine01 (sound familiar?). Towards the bottom, we see several mentions going into one single user in this group, ScottatErin so, let’s look at “Scott’s” network.
Here again we see familiar troll users and accounts scoring very high on our bot score model. Expanding out these relationships a bit more, we see a pattern starting to emerge.
We note an interesting spot where there are a number of retweets (blue lines) from one hub, that connect to a number of mentions (green lines) from another hub.
These two hubs correspond to retweeting Tweets from jonkay (an influencer of sorts hostile to the Canadian liberals) and mentions of MaximeBernier. Looking at the 33 users that retweet jonkay and mention MaximeBernier, the largest node is MongrelGlory, an account with serious #MAGA, #KAG, and Biden-bashing credentials. If you look at the followers of MongrelGlory, you quickly descend into the realm of bots and trolls. It is a spiral of misinformation and fake personae that is consistent with every other phase of this research.
In this work, we keep finding Twitter user_descriptions like this in the networks of accounts promoting Bernier and attacking Trudeau:
“🇺🇸 I ❤️ & PROUDLY SUPPORT our PRESIDENT TRUMP & COUNTRY 🇺🇸#MAGA 🇺🇸 #TRUTH 🇺🇸 #HONESTY #HOPE #FAITH #PEACE #FREEDOM 🇺🇸🇺🇸 LEGAL U.S. CITIZENS 🇺🇸🇺🇸”
Again we have to ask: when did #MAGA folks get so interested in Canadian politics?