Canadian Elections 2021 - Real Time Sentiment

Please use a desktop or laptop to view this webpage as it is not formatted for mobile devices

Want more in-depth analysis or more frequent refresh cycles?

Disclaimer

 

  • Sentiment should not be mistaken for 'advocacy', 'quality', or 'opinion'.  It simply means that the predicted view or attitude towards the Tweet is either positive or negative 

  • The sentiment analysis model is based on an open-source and pre-trained machine learning model available on Hugging Face (see model details below).

  • Think Luna is not responsible for the accuracy of the model's predictions, nor takes any liability for decisions or outcomes as a result of viewing or consuming this dashboard or the data within.

  • We are not affiliated with, nor advocating for any political candidate or party - no candidate or party is intentionally favoured or disfavoured. The data is presented as it is collected. We ask you as the reader to do further research before forming opinions of any party or candidate. 

  • This dashboard has been created purely for educational purposes and is not meant to be a substitute for polling

Model Details

 

Provided below are some high-level details of the model used to develop this dashboard. Stay tuned for our blog post on an in-depth review and analysis

NLP / Sentiment model used: DistilBERT base uncased finetuned SST-2 

How 'Overall Sentiment' is assigned:

  • If Quoted Text is blank and the Retweeted Text is blank, then the regular Text field's sentiment is used

  • If Quoted Text is blank and the Retweeted Text is not blank, then the regular Retweeted Text field's sentiment is used

  • If Quoted Text is not blank and the Retweeted Text is blank, then the regular Quoted Text field's sentiment is used

  • If Quoted Text is not blank and the Retweeted Text is not blank, then the regular Retweeted Text field's sentiment is used

  • In all other situations, the regular Text field's sentiment is used

How we determine the party:

  • We specify some keywords associated with each party.

  • If a keyword shows up in any of the text fields, that party is assigned

  • Multiple parties can be assigned to one Tweet

How we measure '% Positive': For each party, Tweets with Positive Sentiment / Total Number of Tweets

Stay tuned for our blog post where we go into more details of what keywords we used. 

Related Articles