Most of us are familiar with recommenders. Recommenders were first implemented in the early 1990s by applying the idea that sifting through information was a collective rather than individual activity.(Riedl) You are likely to be familiar with the recommender system on Amazon.
Many recommender systems work by using a process called collaborative filtering. Collaborative filtering works by comparing your likes and dislikes to the likes and dislikes of others. The collective data can then be used to extrapolate a new potential like for you based on the likes of other people who you compare well with. If you think about it, collaborative filtering is an old idea. If lots of your friends like a movie and you usually like what they like then there is a good chance that you'd like the same movie. What's new is that collaborative filtering can be automated with computers using data from thousands of users instead of your handful of friends.
Users may not realize it, but they need recommenders. One study shows that Internet users are using search engine sites as answer engines and are merely "[grabbing] query-related nuggets from sites, but ...[not engaging] with the sites themselves."(Nielsen) As an internet user, I would rather have customized recommended content delivered to me instead of reading the digital equivalent of chit-chat.
Users may have fears about the user of recommenders. In order for a recommender to customize results for you it must know who you are. This means somewhere, digital data about you needs to be collected and stored. I attended the World Usability Day event here at Michigan State University in November of 2005 and a question concerning this fear was brought up by someone from the audience after a presentation by Joseph Konstan, an expert in recommenders. It was particularly mind-opening when Professor Konstan answered such concerns with a wonderful example that I paraphrase from memory below:
Maybe a generation or two ago, it was not uncommon to go to a bookstore where the clerk knew you very well. He may have told you not to bother looking around because he has two books behind the counter which he set aside with you in mind. People used to have to shop in their own neighborhood where everyone knew you and knew what you liked. The use of computer recommender systems can recreate that form of excellent customer service for our virtual global neighborhood.
It is likely that recommenders may need a PR campaign before some users fully accept them, especially with our current political climate and the increase in the crime of identity theft. Others have fully accepted recommenders so much that they are willing to pay for the recommendation subscription service TiVo.
I do agree with some of the public fear over data collection and privacy issues. However, I feel that this sacrifice may be worthwhile when recommenders can provide the freedom to sail easily through the current sea of information rather than be flooded by it. Eventually, the plan is to move beyond user-initiated behavior and have the computer provide just-in-time information before the user makes a request.(Hammond) For example, interfaces would know to open up a word processor without you actively requesting it. When computers can predict requests, it won't be long before your refrigerator becomes a computerized domestic servant and knows not only what groceries to restock but also what new exotic fruit and recipes to suggest you try. With such technology, you can just sit back and let your own customized computer crew navigate and deliver you directly to your destination.
Read an entertaining story about a computerized domestic servant.
There are many benefits for management to adopt recommender systems for their sites. Beyond individualized customer service, management will want to use recommenders to determine which products to stock or which development areas to pursue. Management may also be able to use recommenders to improve affiliate relationships. Companies could combine their data with the data of their affiliates and exchange both advertising and customers with each other. Users would likely not feel like they were being sold to because your product would be discovered through a recommendation rather than an advertisement.
Developers need data from users for their recommender systems to work with. A new site may not have enough data early on to extrapolate any predictive trends. The recommender MovieLens gets over this hurdle by forcing all new users to rate a set of movies. This tactic may not be practical for the site you are developing. Users may need an incentive to provide ratings. Management and developers will need to work together to come up with ideas for virtual money to pay customers for their opinions. For example, a store could pay users with coupons and a product site could pay users with free a warranty extension. Developers may be tempted to follow Amazon's lead and ask users to create wish lists and listmania lists without paying their users with virtual money. For those developers who want to create a new method for collecting user data management may still encourage developers to simply copy whatever Amazon does. Developers and designers need to realize that their store or product is probably different than Amazon so they should research what method best suits their specific needs before implementing a system.
is a gradute student in Digital Rhetoric and Professional Writing at Michigan State University. She is also part of MSU's support staff in the Division of Science and Mathematics Education where she works as the support coordinator for the MSU-developed course management system LON-CAPA.
berryma5 AT msu DOT edu