Predicting Information Cascades in Social Media
An information cascade is, in essence, the tendency of people to make the same decisions as the people they see around them, in a sequential fashion.
Most social media websites today allow for content to be shared and re-shared with the click of a single button: Twitter has retweets, Facebook has the share button, and many other platforms have their own version of the same feature, allowing users to share the posts they find interesting with their friends or followers within seconds. Because the spread of ideas can happen so quickly and on such a large scale, it is possible to observe information cascades in action on social media. But can we predict them?
Turns out that while it’s pretty difficult to know for sure which posts are going to go viral as a result of cascades and which ones aren’t, there exists a way to predict it with impressive accuracy. In a paper published in 2014, Justin Cheng and his team studied the way media (specifically images) were shared on Facebook over a period of 28 days after being uploaded. This included 150,000 images which were re-shared over 9 million times total. By plotting the network of each user that shared each image, the researchers were essentially able to construct the flow of the information cascade by tracing the flow of information from user to user. Using this information to calculate the likelihood of an image being shared twice as many times as it has already been shared, Cheng’s team was able to identify a set of traits that correlated to higher likelihoods of images going viral: Whether the original poster has a large amount of followers and whether the post includes a caption among others. The most crucial factor appears to be the initial speed at which the image is being shared around, meaning that if an image starts spreading quickly, it is likely that it will continue to be re-shared across Facebook. The accuracy of the predictions improves with the number of shares an image has received, as more data is available.
While this paper focuses specifically on Facebook shares, it can be useful in informing how we think about information cascades in a more general sense: If a kind of trend or behavior is spreading quickly among our social circle, it’s more likely that we will also participate in it, thus spreading it to others in the process, continuing the cascade. This can be in terms of choosing to conform to a fashion trend, following a political belief, shopping at a certain store, or many other decisions in our daily lives.
References:
https://arxiv.org/abs/1403.4608#