Google's Clustered Web Results
Web Design & Technology News, March 9, 2005

March 9, 2005

SAN FRANCISCO--Google's 8 billion-plus Web document index may not multiply, but its search engine will learn to better divide the data.
By Stefanie Olsen

That was part of the message from Peter Norvig, Google's director of search quality, who on Tuesday gave a keynote speech here at the Semantic Technology Conference. Norvig, a former NASA employee and author of books on artificial intelligence, highlighted several research projects the company is developing to help classify data and improve the relevance of search results.

Those projects focus on adding new clustering capabilities for search results, providing suggestions for related searches, personalizing listings, and mining factual answers to answer queries, Norvig said.

"We want to have a broader bandwidth for that kind of communication," Norvig said. "It's a question of what's the right language."

Despite heavy competition in recent years to own the largest document index, Norvig also said he couldn't foresee Google's database adding much more Web documents without cataloguing bogus or useless pages. Still, the company has numerous programs to add otherwise inaccessible data, like that from books or TV shows, to its Web search engine.

Norvig highlighted a research paper a Google employee wrote last year for a classification engine, which the company is testing. The technology can parse a proper noun or compound nouns into several categories in order to deliver clustered results, for example. For a query on "ATM," the engine would be able to use the terms "such as" on indexed Web pages with the term to discover that it can be linked to the term "high-speed networks." As a result, a search for high-speed networks, might pull up a cluster on ATM.

Norvig said the same technology could be used to mine factual answers from the Web for queries like "President Lincoln's birth date." The technique could have an edge over Microsoft's recent addition of encyclopedic answers to its database, thanks to its Encarta software, Norvig said. That's because MSN's engine could miss the chance to deliver the desired factual answer if the searcher's query is inexact. In contrast, Google draws on the semantic Web and various language sets from pages to find a match.

Norvig also demonstrated Keyhole, Google's satellite mapping service. He said that over time, the company would greater integrate its maps and local information on businesses and sights. "It's important to deliver information about the real world as people carry devices around."