NEWS: New Search Strategy Untangles the Web

Science  01 May 1998:
Vol. 280, Issue 5364, pp. 647b
DOI: 10.1126/science.280.5364.647b

At a time when Internet traffic is doubling every 100 days and searches can haul in tens of thousands of Web pages, researchers are scrambling to find new ways to pare down the info. One promising new approach is to analyze the hyperlinks connecting Web pages—not their text—to pick out the most useful sites for a topic.

Among those pursuing the idea is Cornell computer scientist Jon Kleinberg and colleagues at the University of California, Berkeley, and IBM. When you put, say, “jaguar” into the search engine Alta Vista, Kleinberg notes, you'll get 440,000 pages on everything from jaguars in zoos to car dealers and the Atari Jaguar computer. His group's program narrows this list by assigning a weight to each page according to its links to and from other sites; lots of links toward a site make it an “authority,” Kleinberg says, while pages with links to authorities are “hubs.” After several iterations, the hubs cluster around the authorities, he adds. The result is four or five sets of perhaps a dozen sites, each set about just one topic—for example, the Jaguar computer.

“In a lot of cases, [each set] is comparable in quality” to the best Web resource lists compiled by people, such as Yahoo!'s lists, says Kleinberg. IBM has applied for a patent on the algorithm. Look for links-based searching strategies from a half-dozen other universities pursuing the idea, Kleinberg says, as well as from search engine companies.

Navigate This Article