Gervacio Essay 2

Aimee Gervacio
Knowledge Management
Essay 2
May 12, 2010

Recommender Systems and Knowledge Management

Introduction

Recommender systems have begun to integrate themselves into today’s mainstream media and Web 2.0 technologies. Pandora Radio is an automated music recommender system that uses the Music Genome Project to give their users song selections that is musically similar to their artist or song choices. Netflix uses a video recommender system that, like Amazon’s own recommender system, suggests movies for its users based on user reviews and ratings. StumbleUpon allows users to rate and review websites, allowing new users to find websites that have been recommended by other users. ITunes’ Genius feature is yet another music recommender system that creates a playlist based on that user’s library of music. All of these are just a few examples of not only moderately to extremely successful recommender systems, but also attests to the idea these systems have a two-pronged approach to users’ knowledge management. The first is that these systems are a service to the user; they help the user find artists they might like, a movie that might interest them, a book they could read, or a website that they can visit again in the future. The second prong is that these systems are also an effective form of advertising; users can discover new artists, movies, books, websites, etc that they can purchase. While there are many companies with successful recommender systems, none, at the moment, can compare with the recommender prowess of Amazon.com.

Amazon.com

Why It Is Good

Amazon.com was founded in 1995 as an online bookstore; according to their 1997 letter to the shareholders, they “brought [customers] much more selection than was possible in a physical store…and presented it in a useful, easy-to-search, and easy-to-browse format” (Amazon). This is still true today; browsing through the website is easier than other e-commerce sites. Since then, it has blossomed into one of the largest electronic commerce companies, offering digital media, clothes, and much more. They “seek to be Earth’s most customer-centric company for…consumer customers, seller customers, and developer customers” (Amazon). Customer satisfaction is important to their business. One of the biggest factors for why Amazon has become so successful isn’t low prices, large selections, or easier browsing; it’s their recommendations system.

According to a 2003 article written by Amazon employees Greg Linden, Brent Smith, and Jeremy York, Amazon uses recommendations algorithms to “personalize the online store for each customer” (Linden et al). What’s more, their algorithm “scales independently of the number of customers and the number of items in the product catalog” and creates recommendations “in realtime [and] scales to massive data sets” (Linden et al). The article goes on to describe traditional collaborative filtering and compares it with Amazon’s own item-to-item collaborative filtering system, which takes a user’s purchased or rated items and finds similar items. For example, if I purchased a book on how to cook, my Recommendations page would reflect similar books to that purchase. Or if I rated a book, Amazon would recommend books by that author or other books with similar writing styles and genre.

By focusing on what I have purchased or rated, Amazon is able to provide a service to me by recommending similar products that I might be interested in, exposing me to new products. It can also promote the sales of said similar products, which boosts its advertising sector. Because Amazon cannot give me recommendations with me giving them data on products I purchase or rate, the Amazon recommender system becomes an effective source of my knowledge management. What I get out of it is only as good as what I put in it. This can be productive or, unfortunately, counterproductive.

Why It Is Bad

How is Amazon’s recommender system counterproductive? Why can it be bad? Because what I get out of it is only as good as what I put in it. So while Amazon keeps track of everything that I purchase, review, rate, or browse through and gives me recommendations based on that information, it can also give me unnecessary recommendations. For example, I, along with many other college students, tend to buy my class textbooks online because it’s cheaper than buying my books at local bookstores. While Amazon gives me many cheap choices to pick from, it also makes recommendations based on what I purchase. If, for example, I had to take a class in medieval literature and bought my book from an Amazon retailer, many of the books that would appear in my Recommendations page would be medieval in style and genre. If this was a one-time purchase and I had absolutely no interest in reading any more medieval literature, Amazon has no way of knowing that unless I specifically click on “I don’t like it” on the ratings page. It’s not an easy process of getting a one-time purchase out of the Recommendations page; in fact, it’s near impossible because it can, and will, come back. As a result, it makes the Amazon browsing experience less enjoyable because not only would I have to go through the numerous recommendations and dislike them all, but it also puts me, as a user, on edge the next time I look up a product on Amazon. The ease of finding affordable products becomes secondary to this awareness of what I purchase or rate will affect my recommendations page. Consequently, users would have to be actively managing their Amazon profile, which can be time-consuming and frustrating. The whole purpose of Amazon recommendations as a service and an advertisement is to expedite the online shopping experience for users, not to drag it out. So what can Amazon do?

Genetics and Recommender Systems

Tim Westergren, founder of Pandora Radio, and thousands of other musicians and technologists came together “with the idea of creating the most comprehensive analysis of music ever” (Pandora). Since 2000, they have assembled nearly 400 “musical attributes or ‘genes’” of thousands of songs to determine the “unique and magical musical identity of a song” (Pandora). What this technology does for Pandora Radio is that it broadens the musical horizon of all of its users. Because each song has its own attributes, a user can enter in a song or even an artist into the Pandora search engine, and what Pandora will do is create a station based on the attributes of what the user has entered in. For example, if I wanted a station based on Jason Mraz, not only will Pandora give me all of the songs that Jason Mraz sings, but it will also give me songs of other artists who have similar attributes. It sounds similar to the Amazon recommender system, but Pandora and the Music Genome Project are more exact as to what the user wants, and in a less time-consuming manner. Going with the Jason Mraz example, Pandora will find artists who use also uses acoustic guitars, has live performances, and collaborates with other artists. From there, I can listen to what Pandora has recommended for me to listen to based on my criteria. What I can then do is like or dislike a song that comes up in the station; not only is it simple, but effectively lets Pandora know that, while I like the attribute that is shared by Jason Mraz and this song, I don’t like the other attributes of the song. Pandora gives me and all of its users a broad spectrum of recommendations based on one criterion, but we the users have the power to narrow the field.

What does this mean for other recommender systems? For Amazon, having a set of attributes for its products can prove to improve its recommendations. Books and movies, for example, would be a good place for them to start. While books and movies are already divided into genres, which are then divided into subcategories, having a list of each book’s attributes can help narrow the recommendations. This, in turn, can make disliking an Amazon product into more of a constructive process. As a result, a user’s knowledge management would be under more control and be used more effectively.

Works Cited
Amazon.com. “2009 Shareholder Letter.” Web.

—-. “Investor Relations: Corporate Mission.” Web.

Linden, Greg, Brent Smith, and Jeremy York. “Amazon.com Recommendations: Item-to-Item Collaborative Filtering.” IIEE Computer Society: Industry Report. Web. Jan/Feb 2003. [http://www.win.tue.nl/~laroyo/2L340/resources/Amazon-Recommendations.pdf]

Pandora: The Music Genome Project

Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License