This circumstance prompted the deployment of SAMbot to track toxic tweets received by incumbent candidates and party leaders during the 2021 federal election. As a machine learning bot, SAMbot’s scale is powerful — it can review millions of tweets and distinguish between rude, threatening and sexually explicit content. For more methodology details, visit our About SAMbot page.
Here we present an overview of the toxic sentiment received by 298 public Twitter accounts monitored over the 36 day election period (August 15, 2021 at 12:00 a.m. ET to September 20, 2021 at 11:59 p.m. ET).
For the 2021 Federal Election, we also released weekly SAMbot snapshots. You can view them here:
SAMbot uses a natural language processing, machine learning model to assess toxicity. These models are ever-evolving and thus, everytime SAMbot is deployed it becomes more accurate and informed. Learn more about how SAMbot detects toxicity.
Overview of Findings
Of the 2,550,098 tweets analyzed:
144,177 were labelled as containing sexually explicit content
200,785 were labelled as containing profanity
212,059 were labelled as containing identity attacks
291,393 were labelled as containing threats
421,141 were labelled as containing insults
*Tweets can possess more than one toxicity attribute at a time (e.g. A tweet can be analyzed as both an insult and a threat simultaneously).
Party leaders received a majority of all tweets during this election (70%). Together, party leaders also received a majority of all toxicity (approximately 70% of all likely toxic tweets).
Prime Minister Justin Trudeau received the highest volume of toxicity of any candidate with over 200,000 likely toxic tweets tracked during the election period. Over 1 in 5 (22%) of all tweets received by Trudeau were likely toxic. Trudeau alone received 42% of all likely toxic tweets.
While the volume of toxicity received by Green Party of Canada leader Annamie Paul was low compared to other party leaders. Proportionally, Paul received the second highest rate of toxicity of any party leader with over 1 in 5 (21%) of all tweets received being likely toxic.
Annamie Paul was also the only party leader to receive Identity Attacks as the most common form of toxicity. Identity Attacks were among the top three most common toxicity attributes received by Jagmeet Singh.
SAMbot tracked 88 female identifying and 210 male identifying candidates.*
Female candidates were more likely to receive toxicity than male candidates. Approximately 21% of all tweets directed at female candidates were likely toxic while 18% of tweets directed at male candidates were likely toxic.
Female candidates were more likely to receive sexually explicit comments compared to their male counterparts.
Of the 88 female candidates tracked, 50 were from the Liberal Party of Canada and they received 82% of all likely toxic tweets directed at female candidates. Of this group, high levels of toxicity were tracked for Chrystia Freeland and Maryam Monsef who each received approximately 20% of toxic content sent to female candidates.
Freeland received a high amount of toxicity from August 22 to 23, 2021 when she posted a tweet with a video conveying Conservative Party of Canada leader Erin O’Toole’s stance on healthcare. Twitter labeled the tweet as “manipulated media”, meaning that Twitter viewed the content to be “significantly and deceptively altered” to portray O’Toole in a way that contradicts reality. During this two day period, Freeland received 2,897 likely toxic tweets.
*Gender information was sourced from the Library of Parliament.
High Toxicity Moments
September 7 to 9, 2021 were the most toxic days of the election period. On the evening of September 6, Prime Minister Justin Trudeau had gravel thrown at him at a campaign stop in London, Ontario. The following three days were the most toxic of the election period.
September 7: 20,395 likely toxic tweets were tracked – the highest volume of likely toxic tweets in a single day during the election period. A coordinated bot attack targeting eight candidates also occurred on this day.
September 8: 15,434 likely toxic tweets were tracked on the day of the French leaders debate.
September 9: 14,191 likely toxic tweets tracked on the day of the English leaders debate.
Coordinated Bot Attack
SAMbot can detect when there is a high volume of organized tweets in a short period of time with duplicative language. This indicates the coordinated use of bots to send tweets.
A coordinated bot attack was monitored from September 6 to 8, 2021. During this period there were at least 116,761 tweets (approximately 36 tweets per minute) with content about the civil war in Ethiopia’s northern Tigray region. These tweets were received by eight of the federal election candidates being tracked by SAMbot.
The coordinated bot attack affected the results of different candidates to varying degrees.
Tracked Candidate Overview
Read the Report
Find more detailed analysis about candidates, party and toxicity in the 2021 federal election in our report.
How to cite: The Samara Centre for Democracy, SAMbot 2021 Federal Election Report, (Toronto: The Samara Centre for Democracy, 2022), www.samaracentre.ca/articles/sambot-2021-federal-election-snapshot