Diffusion of False Information: A Good Reason to Fact-Check
This article describes the outcomes of a study relating to the spread of information in what they call “cascades” of retweets on Twitter. As defined by the author, a cascade occurs when novel information is tweeted by some individual and it slowly gains popularity, resulting in lots of other users spreading the same information out of interest, surprise, or perceived importance, regardless of the information’s reputability. Often, the initial tweet contains misinformation (e.g. false information tweeted about the Boston Marathon bombing), and other times the rumor spread is true (e.g. tweets surrounding discovery of the Higgs-Boson in 2012). The author finds that, generally, false information cascades are more common and spread farther than true ones, and often begin with a real user (not a bot) that has a small number of followers. This conclusion is somewhat alarming, since it suggests that any individual user on Twitter has the power to spread potentially harmful false information far and wide, presenting a good reason to fact-check anything one reads online. In this study, the author uses unanimous approval among fact-checking sites like Snopes.com and Politifact.com as grounds for the truthfulness of information in studied cascades, so theoretically using several of these sites is a good way to check claims read online.
The findings of this study are closely related to what we have discussed about diffusion of information in class. A Twitter user is effectively a node in a graph connecting users that see each other’s tweets (everyone initially choosing B), and they may decide to retweet new information (choose A) or not to retweet (i.e. stick with B). The author of the article conjectures that false rumors spread more readily because of novelty, as it is easier to produce novel information if it does not have to be true. This can be interpreted as greater value/novelty in choosing A, i.e. a larger value a that results in a lower threshold fraction of neighbors choosing A for some other user to coordinate. As mentioned in class and on the current homework, a larger value in choosing A makes it easier to spread, which is possibly what the author observes in this study. Furthermore, a user with a small number of followers is intuitively more likely to have followers with fewer neighbors, which increases their fraction of neighbors adopting A from the initial tweet, making their fraction more likely to be above the threshold. This could help explain why false cascades in the article begin with real users that have few followers. The spread of information on social networks like Twitter is thus a good application of recent class material, and also a good reason to be wary of anything read online.