Leveraging News Bias Ratings for Twitter Bias Ratings
As conversations about the 2020 Democratic primary come into full swing, we will track candidates’ coverage much as we did for 2018 congressional races. And we’ve been re-tooling so we can track what might be called the Twitter Primary. By expanding the monitoring framework to Twitter1, we can keep a balanced view of relative candidate appeal, viral stories (whether deceptive or not) and network propaganda efforts.
We began tracking twitter mentions of potential Democratic POTUS candidates on November 13, 2018. For that half day of tracking, we sampled once every hour for a small number (n<10) of obvious candidates.
Note that the top-mentioned “candidate” from that day, Deval Patrick, has since opted out of the race. We should not assume that raw mention count is a direct measure of attitudes in the electorate. But counting these tweets allows a better understanding of topics and trends; and collecting them provides a source of data for better understanding political attitudes, and how they can spread virally.
Understanding the Conversational Terrain
So we also wanted to understand who is talking about the candidates. For the purposes of understanding viral deception and network propaganda this is at least as important as tracking the mentions themselves.
The News Alerts Analogue
Recall from the previous post that we created a two-dimensional visual space to represent the political leanings and relative credibility of various online news sources and news-related sites. In this space, the distribution of coverage during the final week of the 2018 Texas Senate race would be represented as:
Extending the Metaphor to Twitter
Since we do not have Political Bias and Commitment to Fact ratings for Twitter users2, we approximate these coordinates from the link behavior of the users themselves. A twitter user’s coordinates are inferred as an average of:
- The coordinates of the sites that they circulate as links, if they circulate media links; or
- The coordinates of the twitter users that they interact with otherwise
For November 13, we get:
Most of the conversation is credible center-left (i.e. top-left quadrant), but we see a smattering of extremist contributions seeping into the conversation.
Clicking on a few examples in each of the quadrants anecdotally confirms the approximate ratings. For example, the two extreme bottom-right examples in this sample use words like MAGA, KAG and patriot and use flag, eagle or QAnon imagery in their profiles. And common in the top left are “Resistance” accounts circulating news stories from mainstream national news sources.