Perceptions Of Purchasing Patterns
Often overlooked in the maturation of Web technology is the maturation of the Webvisitor. E-shoppers now expect to be shown quickly what a site offers. A number ofcommerce
sites are providing featured items on home pages that can be purchased with a click.Web users are also willing to trade identification and profile information for timesavingrecommendations. Both characteristics are perfectly suited for collaborative filtering.
IN THEIR RIGHT MINDS
In collaborative filtering, a visitor is identified as part of a group of like-mindedpeople and provided with a selection of highly rated items they haven't yet evaluated. Theresult is a word of mouth, serendipitous sales opportunity. The results of a collaborativefiltering question can at times produce almost laughable results: Do I really seem likeI'd like Christopher Lambert's movie Mean Guns? But the very simplicity of its designensures that it remains reasonably close to the mark.
As MIT researchers Rob Guttman and Daniel Dreilinger stated at a seminar on agentsoftware, traditional feature-based comparison software "often finds useless thingsnot sought for" whereas collaborative filtering "often finds useful things notsought for." Collaborative filtering can play a role in the sophisticated world ofmarketing Consumer Buying Behavior models.
Technology vendors such as Net Perceptions, LikeMinds, Jango and Firefly Network (aMicrosoft acquisition) have begun enabling commerce Web sites with collaborative filteringengines. Formerly known as GroupLens, the Net Perceptions For E-commerce recommendationengine can be plugged into Web catalog products such as Microsoft Commerce Server,BroadVision's One-to-One, and IBM Net.Commerce, and has been integrated directly by firmssuch as Amazon.com, CDNow and the iVillage Woman's Network.
LikeMinds technology has been integrated directly by merchants with highly custom sitessuch as Columbia House, Home Box Office, Cinemax and West Coast Entertainment.
SHARING A VISION
BroadVision is a Net Perceptions reseller (and HP Partner), offering its recommendationengine as one of the One-to-One Intelligent Mapping Agents. By implementing a MicrosoftActive User Object provider (through ADSI), Net Perceptions for E-commerce can be easilyconfigured to drive the Personalization & Membership feature of a Microsoft CommerceServer catalog. Net Perceptions' Ad Targeting product can similarly be plugged intoMicrosoft's Ad Server to boost a banner ad's click-through rate. (Microsoft CommerceServer itself provides a simplified collaborative filtering component, the IntelligentCrossSell Predictor, that makes recommendations from purchase data.)
Outside the storefront, Vignette's StoryServer content management oriented Web servermakes content recommendations with a built-in "express" version of NetPerceptions' engine. And Oracle's FrontOffice Marketing software comes standard with a NetPerceptions engine.
While such integrated solutions exist, there is often a meaningful return associatedwith customizing the interface to a sophisticated recommendation engine, especially in thearea of collecting preference information. The ways of gathering preference informationfrom visitor behavior can be as varied and unique as each site's shopping or visitingexperience. Yet, the more preference information an engine has to work with, the betterthe performance of resulting referrals will be.
At the SEPIA Video Guide (vguide.sepia.com), you can see collaborative filtering inaction yet still remain fairly anonymous while remembering and enriching its knowledge ofyour film preferences and perhaps reconsidering some decent films for rental.
Sites with repetitive purchasing of products, such as rentals, books, music andinformation are prime candidates for collaborative filtering. Many current implementationsrecommend commodity products. That's why collaborative filtering is appealing to onlinebuyers and is poised to become a very visible part of the Web experience.