This blogger skimmed the ebook, “Word of Mouse” by John Riedl and Joseph Konstan, published in 2002.
It is about the concept called “Collaborative Filtering.” That means the ability to make product-recommendations to consumers based on a significant number of their self-reported likes of products via an algorithm in a computer program.
The authors claim that the program makes recommendations with a high degree of accuracy, once a subject provides sufficient data on likes and dislikes. Such data are superior to demographic data such as age, occupation and sex, when it comes to predicting future preferences.
Collaborative filtering can be applied to sales of clothing, books, movies and goods sold on the internet– simple products that are purchased according to taste. “Cultural tastes seem to run in patterns.”
This blogger theorizes that the algorithm would do poorly on complex offerings that involve customer service– restaurant meals, hotel rooms, flights or personal services, because they are an experience that varies every time and are more likely to be enjoyed multiple times. A singular product like a book or movie, is a one-time experience.
When polled by the computer program on a book or movie, consumers express their like or dislike only for the book or movie, not bookstore atmosphere or moviegoer rudeness. Consumers might rate a hotel room on hotel-staff friendliness, room decor, cleanliness, and a host of other variables; if they have stayed at the hotel more than once, the rating might also reflect consumers’ general vibe about the hotel for all their stays. On any given day, the consumer might have a good or bad experience at a hotel. Anyway, the algorithm might achieve the same degree of accuracy by recommending a hotel simply based on other hotels with similar amenities and features, as by recommending based on the consumer’s likes of other hotels.
The authors discuss an online business that was started in 1998, Priceline, which allows customers to name the highest price they are willing to pay for a product or service, and if their purchase is approved, (presumably) receive it at a deep discount. For the most part, this appears to be irrelevant to collaborative filtering. Nevertheless, interestingly, the “reverse-auction model” has turned out to be profitable for travel-related services but not for gasoline, groceries and financial services. The reason is that airlines and hotels suffer a total loss on each plane seat and hotel room unfilled on any particular flight or night, respectively. Recouping some revenue from passengers and guests, even at a deep discount, is preferable. The authors make a point about how Priceline displays local geographic expertise in selling its services. Displaying expertise is important for online selling.
The authors boldly proclaim, “We envision recommenders moving out more into the public and the bricks-and-mortar sphere… Recommenders can limit the number of items a customer needs to see on each [Web]page… Recommenders can also be used in voice interfaces where the limiting factor is low bandwidth…” Clearly, Riedl and Konstan underestimated the algorithmic proficiency of Google.
Read the book anyway to see the authors’ enthusiasm for collaborative filtering and get numerous tips on online selling, marketing, and what we now know about the internet. 🙂