Pedro Domingos and Matt Richardson analyze the power of viral marketing in their paper, "Mining the Network Value of Customers." Popularized by The Tipping Point, viral marketing exploits the power of word-of-mouth networks by targeting customers who will help promote products to their friends and colleagues.
Put another way, viral marketing relies on the insight that each potential customer has not just an intrinsic value but also a network value. The intrinsic value of a potential customer is the expected profit from sales to her (based on likelihood that she will buy and the marketing expense involved). The network value of a potential customer is the expected profit from other customers she may influence to buy, both directly and indirectly. This idea is probably old hat to virtually all of my readers, who, like me, are well entrenched in the "network value" camp.
What I found interesting about Domingos and Richardson's paper is their quantitative approach. Rather than merely arguing that networks are important, they instead ask exactly how much is network value really worth? And how can viral marketing best exploit that value?
To tackle such concrete quantitative questions, the authors not only develop some heavy mathematical machinery, but they also propose an interesting source of concrete networks to test their models on. Here again I find their approach thought provoking. (As I discussed in a recent post, finding an actual network can be the hardest part in this kind of analysis.)
The authors use collaborative filtering databases to build their networks. These databases are the foundation of any website that offers recommendations to its members based on their previous recorded purchases and the purchases of other members with similar profiles. Amazon is a famous example of a very effective use of this technology.
Once someone (Jill, say) starts buying books on Amazon, the website tracks her purchases and compares her buying history with everyone else using the site. As Domingos and Richardson point out, this implicitly defines a network, where Jill is linked to the members with buying histories most similar to her own. These are exactly the members whose product reviews and purchases inform the recommendations provided to Jill each time she logs in.
The authors actually use EachMovie, not Amazon. With easy access to the complete history of movie reviews on this site, they construct a network and compute the exact network value of each member of the EachMovie website.
Their results should be encouraging to anyone in the business of viral marketing. They find that a very small number of people have network value of 20, 40, or even 60 (meaning that selling one of these people on a movie implies 20, 40, or 60 more sales as a consequence). Most people have a network value of approximately one.
I am encouraged that the authors have constructed actual networks based on real-world data rich not only in relationships but also purchasing history. However, I do find it puzzling to think of the kinds of network relationships captured through this kind of data mining. When I log onto Amazon, chances are I don't know any of the people whose preferences are informing the book recommendations provided me. And even if I do know them, the recommendations come up anonymously anyway. I am not familiar with EachMovie, but I am guessing that there as well the network is very apersonal -- hardly the image one usually considers in the context of word-of-mouth marketing (aka viral networking).
On the other hand, once I look into a specific recommendation, then I see reviews where not only is the product ranked, but the reviewer is ranked as well. Here is the perfect partner to viral marketing. While viral marketers are seeking those with high network value, many ambitious reviewers are seeking more recognition for their talents as reviewers. If we checked Domingos and Richardson's evaluation of network value against rankings of reviewers, I bet we'd see a pretty good correlation.
For an enlightening take on the world of on-line reviews, see Amazon Secret Weapon No. 1376: the race for recognition on Shorewalker.com.