IEEE Multimedia, vol. 3 issue 3, pg. 10 - 15.
Oct. - Dec, 1996.

What would it take to build a search engine possessing the knowledge, intelligence, and resourcefulness of my favorite research librarian? Her ability to interpret my inquiries, knowledgeably expand them, and then extrapolate them into a rich model for search and retrieval makes her an invaluable and pleasurable resource. She takes pride in knowing her library thoroughly, both spatially (where to physically find a book) and temporally (how her book inventory has changed and evolved). As she formulates her plan of attack, her sophisticated understanding of language, culture, experience, and other knowledge domain models simultaneously converge and are re-mapped onto the reality of her library. The task of information retrieval becomes an interactive adventure of human dimensions, full of the satisfaction of continuous discovery. Keyword matching is a crude and unsatisfying method for sampling the information content of complex sources such as the World Wide Web. These truncated representations fail to model the larger meaning embedded within the source-they leave behind all vital contextual information, and they strip away any basis for appraising the quality and veracity of the source. The worthless appears as the equal of the worthy. In the electronic future, when all books are digitized and available on line, I pray that we have a more skilled and interesting guide than Yahoo or WebCrawler staffing the circulation desk.

Get the paper in PDF or HTML form.