This is the main tool we use internally to verify the accuracy of Emowatch.
When you click the link above, you will be presented with a list of news sources like CNN and BBC. Choose a source and the page will download the last 50 articles from their Facebook page and perform various forms of analysis. This analysis includes:
- Content: the full headline that was downloaded with publication date and time and the numbers of Facebook reactions (love, haha, wow, sad, angry) already given to the headline
- Emotions detected: click the Play button to view a colour coded view of the emotions we predict that readers will feel while they are reading the text
- Predictions: we show the top two emotions, the intensity of the highest emotion and some analysis based on our own research. Where available (when enough people have reacted to the article) we also show the actual results along with our predictions
- Distribution of emotions: a line graph showing the percentage of each emotion predicted for this headline, along with actual data if available
- Distribution of actions: a completely separate model which tries to predict the relative proportion of comments, shares, and likes to this headline, along with actual data if available