InformationWeek
Defining The Business Value Of Technology
December 22, 1997
Just Add People
Collaborative filtering brings human input to information retrieval in the enterprise
iCollaborative filtering builds profiles of end users' interests and recommends content one user finds appealing to other users who have similar tastes. For example, if a user interested in Japanese exchange rates finds a useful Web site on the topic, he or she can rate the site highly and store a link to it in a database. The link will then show up on Web pages served to other users with the same interest. In this way, collaborative filtering supplements stat
istical filtering by adding human judgment.
Information retrieval products in general lack human input, says Jim Bair, an analyst at Gartner Group Inc. in Stamford, Conn. "For knowledge retrieval," he notes, "you need people."
IBM's Intelligent Miner for Text will function like a data mining engine, except that it will mine text instead of numbers. It will combine a full text-search engine with algorithms to analyze language, extract passages of text, and group related passages in clusters. Knowledge Utility-already available to independent software developers to integrate into their products-will add the ability to create profiles and retrieve content that users with like profiles deem important.
Knowledge Utility's real power is its ability to learn individual users' preferences from metadata, IBM officials say. It recognizes several different types of metadata, including types used to describe the contents of Web pages and Lotus Notes documents. Knowledge Utility builds a precise profile of a user ba
sed on metadata about the content the user has accessed from multiple sources. Individual profiles are then compared to make recommendations.
NetPerceptions stresses collaboration. The Minneapolis vendor this month released an upgrade to its GroupLens recommendation engine with enhanced features for small groups. Version 2.5 lets users submit a specific group of people and generate a list of recommendations based on their collective preferences. It also shows a list of all the users who rated an item and how they rated it, or how one user rated an item.
Helpful 'Neighbors'
SceneServer tester Ernst & Young LLP is using it to support distance learning among distributed groups of people who share common interests. The New York consulting firm hopes SceneServer will help instructors move beyond one-way lecturing. Tom Solomon, a senior partner at Ernst & Young, explains: "We're envisioning a learning community consultant who sits back and sees what people are asking for and puts them in touch with the right people and resources."
ollaborative filtering software, commonly used for recommending products on consumer-oriented Web sites, is coming to corporate intranets. IBM next quarter will release an intranet-based text-retrieval product that incorporates the company's collaborative filtering technology, Knowledge Utility. Other collaborative filtering products for intranets shipped this month from startup vendors NetPerceptions Inc. and Digital Knowledge Assets LLC.
Digital Knowledge Assets, in Chicago, has integrated GroupLens into its SceneServer intranet application, which it launched this month. Users submit Web pages and other resources to a database, rate them, and add comments. GroupLens generates a personalized home page for each user showing resources likely to interest them. The page also shows a list of "neighbors" with similar interests and lets users see what their peers are
viewing and how they rated the content.
Featured Microsite
