Using 360° Stance Detection to Analyze coverage of Donald Trump by CNN

Last week, we introduced 360° Stance Detection, a model that classifies the attitude of a journalist toward an entity in a story as either ‘for,’ ‘against,’ or ‘neutral’. Specifically, we talked about how it could be used to combat filter bubbles by recommending news content to media consumers containing different points of view to what they have already read. This week, we’re going to try the model out on a data science use case and analyze the stances taken by CNN and Fox toward President Donald Trump.

You can check out this interactive demo we put together to let you try out the 360° Stance Detection model yourself. Given a topic and a keyword, the demo will source 50 stories about the topic that also mention the keyword and classify the stance of the author of each story towards the keyword. It will then display the results on a scatter plot, plotting the logo of the publisher according to the stance (on an x-axis ranging from ’against’ top ‘for’) and Alexa ranking of the publisher (on the y-axis ranging from ‘less’ traffic to ‘more’ traffic).

For this blog, we’re going to use the model to look into the subject of one of the most popular debates around journalists taking sides in their coverage – that of CNN toward Donald Trump. The President has long maintained that CNN expresses a consistently negative bias towards him, and using the 360° Stance Detection model, we can look into this claim.

Specifically, we’re going to:

1. Quantify the stances taken towards the President in articles published on CNN.com.
2. Look at the stances taken by a specific journalist, Chris Cillizza, Editor-at-Large at CNN.
3. Compare the what we find in the content from CNN with content from Fox News, a source the President publicly praises.

What can Stance Detection tell us about the Coverage of Donald Trump by CNN?

Anyone who pays attention to the news is familiar with President Trump’s claims that CNN continuously bash him in their news coverage, so we decided to use the 360° Stance Detection model to analyze this claim. On the chart below, we plotted the monthly volume of stances detected toward the President in stories from CNN.

We can see that the vast majority of the content from CNN about Trump is neutral, but that journalists also take non-neutral stances toward the President. Some of these are ‘for’, which are mostly those published between Trump’s election victory and his inauguration, although the overwhelming majority are clearly ‘against’. There are clear spikes in the volume of stories with stances ‘against’ the President in months with the following noteworthy events:

• October 2016, the month of the Access Hollywood tape leak
• August 2017, the month of the President’s response to incidents at the Charlottesville protests
• October 2017, the month of his controversial call to the family of a fallen soldier.

What was the stance taken by individual journalists at CNN?

We can also track how individual journalists covered Trump, and how that coverage changed over time. Below, we have plotted the stances detected toward the President since April 2017 in stories by Chris Cillizza, Editor-at-Large at CNN.

Similar to the trends we saw in the in the previous graph, we can see that Cillizza’s stances ‘against’ the President become more frequent in August and October 2017, months containing particularly controversial news cycles.

But not all journalists at CNN are as universally ‘against’ Trump as Cillizza is. For example, take a look at the varying stances the model detected toward the President by Eric Bradner, which we’ve plotted on a graph below:

You can see that Bradner authored some stories that the model classified as ‘against’ Trump in the run up to the election, but he actually published more articles that were in favour of Trump than against him over the Spring and Summer of 2017.

Are the Stances taken by journalists at Fox News much different?

When it comes to Trump’s attacks on CNN, the other side of the coin is his frequent praise of Fox News.

With that in mind, we decided to compare what we found about the stances of CNN journalists with their Fox News counterparts. Take a look below at how the volume of articles that are neutral, for and against compare for each source.

We didn’t expect to find these results. We can see from the chart that authors at CNN take stances ‘for’ the President much more frequently than authors at Fox do. This contradicts what we assumed we’d find, since we’re used to seeing clips of Fox News TV hosts vocally defending the President and attacking his rivals.

We think this is because of two reasons. First, the articles we’re analyzing here are from journalists at foxnews.com, which we found to be quite different from the Fox News TV channel presenters, where the stories seem to be more objective than the television coverage.

Second, Fox’s bias in favour of Trump is visible in a different way – notice how much the volume of stories about Trump on Fox drops in August, the month of the Charlottesville protest. Instead of vocally defending the President, as the Fox News TV presenters did, journalists at foxnews.com simply talked about the President less over this period.

We looked into stances ‘for’ or ‘against’ Trump simply because we were interested in quantifying the volume of stories on each side of a popular debate. If you want to try your own queries to the 360° Stance Detection model, try out this demo that we put together to let you check it out.