Rumors documented by Snopes

We retrieved from the Snopes website two classifications of the rumors they have covered and analyzed. The first is the veracity, which includes “true” and “false”, but also a range of intermediate or orthogonal values, i.e. partly true, multiple truth values, unclassifiable, undetermined, and legend. We also retrieved the broad thematic category Snopes assigned to the rumor, e.g. Politics, Food, “Fauxtos”, etc. After sanitizing the corpus — merging duplicate entries and removing entries with contradictory information — there remained 4,761 distinct rumors.

Around 22% of those rumors are related to politics, 12% are either “photoshopped” images or real photos with fake backstories (Fauxtos) and 11% of rumors fall into a broad category called Inboxer Rebellion which consists of stories of emails and chain letters of “dubious origin and even more dubious veracity”. There is then a long tail of rumors ranging from 9/11 to rumors specifically about Coca-Cola (Figure 1).

Figure 1: Distribution of a number of rumors among cate- gories on Snopes, along with the fraction of these stories that are evaluated as true.

For simplicity’s sake, we recode the veracity in one of three possible values: 45% of the Snopes corpus consists of rumors that have been debunked and are thus considered false, 26% turn out to actually be true, and we place the re- maining 29% (multiple truth values, mixed, undetermined) into a group of rumors that are “maybe” true — either be- cause parts of the rumor are true whereas others aren’t, or because the Snopes team was unable to verify the story. This points to a tendency of Snopes to document more false sto- ries than true ones, in line with the stated mission on the front page of their website: “Welcome to Snopes.com, the definitive Internet reference source for urban legends, folk- lore, myths, rumors, and misinformation