Advertising Cascades
http://cs.iit.edu/~culotta/pubs/culotta03maximizing.pdf
In this study, Aron Culotta of the University of Massachusetts attempts to answer how cascades can be maximized. This follows from the idea that given many innovations in the world, very few will spread overseas while most will be forgotten. He explains that the dynamics of a network itself must influence this spread, and by viewing individuals as nodes in a social graph it may indicate the influence nodes have on one another.
If we imagine a social network containing individual nodes that represent a person in a social network and each directed edge as an indication of a node that influences another, then some graph configurations result in more likely cascades. In other words, an innovation may be more widely adopted even if only a small proportion of individuals initially adopt. This is known as the “global cascade.” A question aimed to be answered is, “Which nodes do we select to maximize cascade size?”
The effect of this can be seen through business and advertising, most notably. Advertising is most cost effective if we choose particular customers to target. This may include things like demographic information, how likely a customer is to purchase a product, and expected revenue a customer may contribute over a lifetime. The study splits buyers into categories such as “always-buy” who will buy a product without advertising, “persuadable” who will buy with a discount, and “anti-persuadable” who will buy only if they are not marketed to (this category, in my opinion, is the most interesting).
However, the most influential effect of an initial purchase is actually a function of the number of previous buyers (similar to our study of cascades in Networks). “Imitators” are highly affected to purchase an invention depending on previous buyers and especially if a customer’s friends have purchased a product. This is known as the “friendship ripple effect”.
Various “solutions” to maximizing cascades are given, including a threshold model and a linear model. However, there seems to be no perfect solution to this. Future work is still to be done to more closely examine effects, view network evolution on cascades, and the many implications this type of study can have on marketing in the future.