A group of researchers from the MIT technology review are attempting to pin down network behavior with respect to information cascades within the confined network space of Facebook. In order to conduct their experiment the researchers (Cheng and his team) took to analyzing the way users shared photographs on Facebook over the course of a 28 day period. They were able to collect data from over 150,000 photos that were shared about 9 million times over their 28 days period, and with that data they were able to see which users shared each photo at what time – all important information for constructing a graph of this behavior.
Cheng and his team note that the typical route to this type of research involves viewing large cascade behavior and attempting to apply some sort of algorithm to predict future cascades. Cheng suggests that their unusual method of watching the cascades grow (compared to looking at end results and working backwards) has allowed his team to better predict cascade behavior. They begin their procedure by observing a single image to be shared with previous shares denoted as K. Then their team predicts how likely that picture is to double its current shares. What they observed in that cascade behavior in this network follow a power law, that half of cascades with a given size will more than double while the others do not.
What they ultimately found given this experiment is that there are given predictors to whether or not an information cascade will occur. They report something called “temporal performance” how quickly the photo spreads and they claim it is the best indicator of all. Cheng and his team explain that “if something is spreading quickly it’s likely to continue to spread”.
And the end of the article they claim, that this experiment although only conducted using Facebook as their subject of experiment, has applications that reach far beyond Facebook and social networking. While I can see how this could make some sense in the real world given their findings, I feel as though there are far too many externalities in the real world for this to work the same way.