cool paper in 3 min (9)

Vosoughi, S., Roy, D., & Aral, S. The spread of true and false news online. Science 359.6380 (2018): 1146-1151.

by Ali Shiravand (*)

“BREAKING: Two Explosions in the White House and Barack Obama is Injured.” This tweet sent out from official account of the Associated Press on April 23, 2013. The news suddenly was spreaded all over the world. False rumor of explosion caused markets to briefly plunge and was wiped out $130 billion in stock value.

The AP quickly said it was hacked but it was too late. Many such online social phenomena bring some important questions to our mind: what is the nature of a false news, how do truth and falsity diffuse differently, and what are the important factors of these differences?

To answer these questions, in this paper Sinan Aral and his team analyzed the diffusion rumours empirically in a large dataset from Twitter between 2006-2017. They examine the behavior of news cascades: a story about a rumor which begins on Twitter when a user makes an assertion about a topic in a tweet. The number of cascades that make up a rumor is equal to the number of times the story or claim was independently tweeted by a user (not retweeted). Each rumor is classified as true or false using information from six independent fact-checking organizations.

Size of a cascade is referred to the number of users involoved in the cascade over time. After analyzing more than 126,000 tweet cascades, they found that falsehoods diffused significantly farther, faster, deeper and more broadly than the truth in the all categories of news (Fig. 1). According to the results, falsehoods were 70% more likely to be retweeted than the truth.

**Fig. 1:** Complementary cumulative distribution functions (CCDFs) of false and true rumor cascades’ size. CCDF shows the probability that a cascade size is larger than any given value identified on x axis. It tells us how often the cascade size is above a particular level. Here we see that cascades involving false news are more likely to be larger in size compared to true news.

In addition, false political news spread deeper and faster than the other categories among Twitter users. This means that people are more likely to share them (Fig. 2).

**Fig. 2:** **Complementary cumulative distribution functions (CCDFs) of false political and**
**other types of rumor cascades’ size. Higher values of false political news means that they are more probable than the other categories with respect to the cascade size.**

I believe that this paper is a very good example of an empirical behavioral study that could help us for better policy-making and emphasize behavioral interventions, like labeling and informing the false news and incentives to prevent the spread of misinformation, rather than focusing exclusively on curtailing spambots. Others have told me that they disagree and the paper does not tell us much about policy. So this is an open question and it would be great to read what you think.

The authors suggest that the novelty of news items is the reason for the different diffusion patterns. This claim is also very controversial. For example, it may be that fake news has stronger emotional content – making the reader more angry, happy etc. Novelty is just one side of fake news.

Finally, it is important to keep in mind that it is difficult to interpret mass behavior from Big Data that has been collected from the internet rather than from a well-defined experimental setup. If you are interested in this topic, Bit by Bit written by Matt Salganick would be very useful for you.

(*) Ali Shiravand is a student at University of Tehran, IRAN where he is doing an MSc in Artificial Intelligence. He also works as a research assistant at the UT Cognitive Systems Lab.