Predicting Information Cascades, and its Implications
https://news.stanford.edu/news/2014/april/viral-photo-cascade-040314.html
This Stanford news article first caught my eye as I was looking through some of Prof. Jon Kleinberg’s research articles and web-mentions. The article summarizes how Facebook researchers, a doctoral student form Stanford and professors from both Cornell and Stanford came together to analyze the “sharing” phenomenon on the social media platform called Facebook. Predicting whether or not a photo goes “viral” or not had always been excruciatingly evasive to data and computer scientists. Prof. Kleinberg and his colleagues took quite an innovative and refreshing approach to break down the problem. Instead of directly trying to solve the “viral photo” problem, they broke it down into a simple basic problem. It was as follows: given a certain number of shares, how likely is the picture to receive double its current number of shares? As Prof. Kleinberg and team came closer to finding the answer to the above mentioned problem, their algorithm improved for when the shares were few. However, the algorithm gave an unprecedented 88% accuracy when run on a great number of existing shares.
This article is very much linked to the topic of information cascades covered in class. Studying information cascades, whether or not they happen, how do they take place, as discussed in class, is very important in the evaluation of social and informational networks. The analysis of this study could also be extended to our categorization of direct-benefit cascades or informational cascades. For example, the reason a person shares a given photo could be either of direct benefit to them, or, as it often is, out of purely informational purposes. The implementation of this algorithm could potentially help us learn a lot of minute and fine details about information cascades on social medias, how likely they are to happen and in what way could they occur. Beyond academic learnings, the insights we receive form this algorithm can be put to multiple practice uses as well. For example, if we can spot a cascade in the making we can potentially also plug an unfolding cascade in the case that it is unwanted. A cascade of fake news, rumors, or violence provoking information can be plugged by the right authorities in time. On the other hand, this algorithm can be used to create information cascades in case of a natural disaster. For example, Puerto Rico ad its citizens could have used some invaluable information about the storm last summer and efficient ways to escape or evade it. Given the influence this research holds over academia as well the private and government sectors, I would love to deep dive into it and explore it more someday.