Information Cascades and Virality on Facebook
Since the dawn of the Internet and especially popular forms of social media such as Facebook and Twitter, the term “viral” has entered colloquial language in a new context. When a video goes “viral,” we know that it means that something went from having a couple hundred views to having several million. What determines whether something goes “viral,” and how does this relate to information cascades and network effects?
To draw a line between these seemingly similar terms, let’s first define what an information cascade is. An information cascade is a situation in which, in a large group of people or users, decisions are made not just on private personal information, but rather on what one sees from others’ behavior. Network effects play a role in social media since the more people use a product like Facebook or Twitter, it becomes more valuable. The more people there are to interact with, the more the product itself is worth a user’s time. Virality, on the other hand, focuses on the rate of the adoption of a product. If users adopt a certain behavior at an exponential rate, then it is considered something that goes viral.
On Facebook, for example, a post is considered to go viral if the rate of the cascade (via likes or shares) increases as more people adopt the behavior (like or share it). In a study of cascades on Facebook by Dow, Adamic, and Friggeri, there are two main ways cascades can be categorized — top-down or bottom-up. The top-down way a cascade can happen is if a popular page posts an image — for example, a meme — which gets many views, likes, and reshares. The bottom-up way would be if an individual user posts something publicly, which at first gets one or two likes/reshares, but speeds up as more and more people share it and it gets more and more publicity.
In the Facebook paper, there were two main cases under study — firstly, a post made by the official Obama Facebook page, and secondly, a photo shared by an individual user. The post made by Obama had a much shallower cascade depth, which means that the majority of people who shared the photo shared it from the root node, or the original page at which the photo was posted. On the other hand, the post from the individual Facebook user had a larger depth, as many of the shares were from non-root nodes (other pages or users sharing the photo). The post made by the individual user is considered as a more “organic” cascade, since the Obama post had other ways that users found the post such as through other websites, news media, etc.
In addition, the time at which a post is made can contribute to the shape and manner of the cascade. Depending on events happening at a certain time of the year, a post can increase in the rate of adoption if the post’s content is relevant to a nearby event. For example, the individual post made by the user spiked near Valentine’s Day even after it become popular because the post had a sexual reference.
Currently, many predictive and classifying models are being made around the topics of social networks and virality. Though a definitive definition is yet to be made around these new topics, this is a very new and ongoing area of research we can look forward to hearing more from.
Dow, P. Alex; Adamic, Lada A.; Friggeri, Adrien. The Anatomy of Large Facebook Cascades. Association for the Advancement of Artificial Intelligence. 28 June 2013. https://www.aaai.org/ocs/index.php/ICWSM/ICWSM13/paper/view/6123