To find reliable outlet articles disproportionately co-shared with fake news, we construct a graph with edges between fake and reliable articles (the nodes) weighted by the number of users sharing both incident articles. For example, the two articles highlighted above, one by a fake news outlet and one by The Washington Post, are both shared by eight unique individuals on Twitter. We then score all pairs of fake and reliable news on the likelihood of being shared by the same set of Twitter users, when controlling for the individual popularity of the two articles (edges in light green (or lighter) shade are below the threshold for being considered as co-shared).
