http://www.cs.washington.edu/homes/pedrod/papers/kdd01a.pdf

Often when companies go about mining for potential customers to market to, they estimate their expected value from a particular customer as the expected profit less the cost of marketing. What the companies do not usually consider is the network value of the potential customer, that is, the expected revenue garnered as a result of other people in the social network of the marketed to customer buying the product. The type of marketing that does explore the network value of customs is called viral marketing, and according to the authors, it “is still a black art.” The intent of the authors’ paper is stated to be to advance the study of viral marketing.

In their introduction the authors state, “Ignoring network effects when deciding which customers to market to can lead to severely suboptimal decisions.” But they go on to say that it is very difficult to quantify the network value of customers. The network value of the customer does not depend solely on him, but perhaps the entire network. In the past, companies have taken losses due to the difficulties of accurately mapping customer networks. Now, however, with the vast supplies of information available via the internet, it has become easier to acquire the data necessary for such modeling.

The bulk of this paper attempts to construct such a model. The authors depicts each customer as Boolean variables (0 – didn’t buy product, 1 – did buy product), and each customer X_i has a set of neighbors N_i, used to describe his social network. To describe their model they use mathematical concepts not discussed in class, but the main idea: to use social networks to better market products, is very closely related to what we have studied thus far in class.