This is based on the co-sharing likelihood derived from Twitter sharing behaviour (specifically, based on an aggregated co-sharing likelihood score for each outward edge from a particular article or URL). As motivation for our hypothesis in this work, consider the third NYT article and the seventh WAPO article shown in the figure about mail-in ballots and absentee voting (which also pop up in our second qualitative case study). These articles are popular in the same Twitter circles sharing stories published by fake news outlets containing narratives (extracted in this work) that include mail ballot raise risk fraud, lot people cheat mail, mail balloting increase incident fraud, paper mail balloting mean dramatic increase fraud and republican complain potential mail absentee ballot fraud. While only headlines are shown here for illustrative purposes, our investigation also used the entire text of the news article. In this figure, we are only showing the top articles that are not opinion pieces, but we note that especially for WAPO, the top co-shared articles set does include opinion pieces.
