The 43rd Canadian general election is on the horizon; it is to be held on or before October 21, 2019. The writ has not yet been dropped, but with seven weeks to go the campaign has unofficially begun and it’s already in full swing online. And whether the discussion and information shared is positive, negative, or outright fake, the state of the debate on social media has the potential to influence voters, media and candidates alike.
As part of the Social Media Lab’s Election 43 (#ELXN43) Transparency Initiative, the Lab has developed PoliDashboard, a new data visualization tool designed to help voters, journalists, and campaign staffers to monitor the health of political discussions in Canada. It analyzes publicly available data from Twitter to spot misinformation and the use of automation (such as the use of political bots).
PoliDashboard consists of three main modules:
- a #CNDPoli Twitter Module that provides near real time analysis of the #CDNPoli tweet stream to flag the presence of misinformation…aka ‘fake news’ being shared on Twitter and the use of automation (aka…political bots).
- an expert-curated Links & Misinformation Database containing all URLs shared on #CNDPoli plus a compendium of blacklisted URLs known for disseminating hoaxes and disinformation (as documented by trusted sources such as journalists & other academic researchers),
- a Facebook Political Ads Module featuring information about political advertisers and the ads they are running on Facebook.
Working together, these three modules of PoliDashboard make use of publicly available social media data to add more transparency to our online political discourse and to put a spotlight on the people and organizations vying for our attention and votes.
With PoliDashboard, a user can answer a wide range of questions from more basic questions such as:
- How many tweets used the #CNDPoli hashtag?
- And how many tweets remain publicly available for viewing after a specified period of time? (i.e., it was not deleted by neither Twitter nor the user)
- What topics are people tweeting about?
- What is the overall sentiment of the #CNDPoli discussions?
To more complex questions such as
- What is the ratio of URLs shared on Twitter link to credible news sources?
- And how many of them link to known purveyors of misinformation?
- What are the most active accounts on #CDNPoli, and are they using some form of automation, i.e., are they bots?
- Who is placing political ads on Facebook, what ads are they running and what is their spending pattern?
As the Election Day approaches, we will provide regular updates from the PoliDashbard with insights and anything that stands out or looks suspicious. Below is our first summary update based on what we observed in #CDNPoli discussions on Twitter posted between August 28 – September 3.
General Stats and Deleted Accounts
During this period, #CNDPoli was tweeted and retweeted 47,357 times – an average of just under 6,750 per day. We’ve noticed that the use of #CDNPoli dropped off on August 29 and 30, going into the Labour Day long weekend. While interesting, this drop off is actually common on Twitter and is not unique to the long weekend.
During this seven day period, based on a random sample of 4,200 tweets, less than 1% (210 tweets) were deleted by the users for various reasons and none were hidden by users or suspended by Twitter. We view these counts as an indirect estimate of how fast Twitter blocks nefarious accounts and posts from its platform. To compare, within the last month we estimated that about 1.5% of tweets disappeared for various reasons and 0.1% of users who previously posted to #CDNPoli were suspended. It is a relatively low number suggesting that either there are not that many nefarious actors on #CDNPoli or that Twitter is not blocking bad actors in a timely manner.

Discussion Topics
Next, we want to know who and what topics people are discussing using this hashtag. Based on the top 10 hashtags co-occurring with #CNDPoli and the most frequently mentioned people, organizations and places, we identified a few insights on what people are talking about. For example, on August 26, the Liberal Party unveiled its campaign slogan “Choose Forward”. Within a week, #ChooseForward ranked among the top 10 co-occurring hashtags, appearing alongside #CDNPoli 794 times. However, #TrudeauMustGo appeared more frequently, being tweeted with #CDNPoli 2,800 times. Also, as somewhat expected, much of the attention is going to the leaders of the Liberal and Conservative Parties. Prime Minister Justin Trudeau was the most frequently mentioned person, while Conservative Party leader Andrew Scheer followed. Our data also shows a figure from the past still maintains a presence. Former Prime Minister Stephen Harper was the third most frequently mentioned person.
Link Sharing
All URLs shared in tweets with #CDNPoli are identified and separated into 11 categories, including mainstream media, government, political parties, partisan sources and known blacklist sites such as imposters, misinformation and malicious websites. (See the complete list of categories and their definitions.) Our data shows that the majority of information being shared is coming from credible sources. Between August 28 and September 3, the majority of web links shared with #CDNPoli were from mainstream media sources (ranging day-to-day between 62 and 59%). Links from partisan sites made up a smaller proportion of information sources (ranging from 10 to 12%). Only a fraction of a percent of domains (57 total) were from identified blacklist sites.



Active Users and Automation
One unique feature of the PoliDashboard is that it identifies the 10 most mentioned users and 10 most active users and their probability of being completely automated (i.e., engaged in bot-like behavior). The most active user was @StradlingDianne, with 2,419 tweets and a 4% probability of being automated. Much of recent activities on the account were retweets and politically related content. In contrast, the eighth most active account @AsAmHollywood1with 1,297 tweets using #CNDPoli had the highest probability of automation of 64%. This account clearly states it is a Twitter bot created for a dissertation project.

Subjectivity and Sentiments
The PoliDashboard also tracks the subjectivity of tweets with the use of opinionated language, and general sentiment, whether it is positive, negative or neutral. Overall, the general sentiment of opinion within the examined period was neutral (38.5%) and positive (34.4%). It will be interesting to track whether the use of opinioned and negative language would increase closer to the Election day.

As October 21 approaches, we will provide regular updates from the PoliDashbard with more insights about the state of political discussions and actors in the Canadian Twitterverse and political advertisers on Facebook, so stay tuned!
By Donald Patterson, Philip Mai and Anatoliy Gruzd.
Press and media inquiries:
Philip Mai
Director of Business and Communications
416-979-5000 ext. 3509
[email protected]