This week, we are covering electronic commerce--or Internet marketing--in my class.
Collaborative filtering is a process whereby an individual's purchases are compared against those of others. The actual algorithm used by Amazon is quite complex, but it basically compares what one customer bought against what was bought by others who shared one or ore of these specific purchases.
A simpler case involves the simple correlation of the purchase of a specific book with purchases of other books. Here, you will see the phrase "Customers who bought this item also bought:" and then a list of the books most frequently purchased by others who have bought this book.
To illustrate, I used the example of the author Jonathan Kellerman, who writes murder mysteries in which a psychologist applies his insight as a consultant to the police. I then off handedly mentioned that I had been toying about the idea of a rather eccentric and absent minded professor--purely fictional despite any possible resemblance to actual individuals--who solved murder mysteries. I then pursued my point: Fiction books can be difficult to classify, so a lot of people might enjoy Jonathan Kellerman books without ever learning that another psychologist--Stephen White--also writes murder mysteries with the protagonist being a psychologist who uses his gifts in much the same way. Using the correlated list, however, it only takes a small percentage of people who know of both for the knowledge to spread readily.
The same idea can be applied to music. What causes artists to be perceived as "similar," or, more precisely, "enjoyable from the same taste perspective?" This could be driven by lyrics, vocal sound, musical style, or a number of other factors. Without knowing the exact classificatory variables, correlated sales can again pinpoint such complementarities.