Home | Discussions | Print Edition | Archives | Site Map | Home Delivery | Advertise | Help
Thursday, October 28, 1999
Home Edition
Section: Business
Page: C-6

The Cutting Edge;

Designing More Useful Search Engines;
Internet: Several firms are mixing human judgment and automation in efforts intended to bring users better-sifted information.;


A new trend in search engines that attempts to blend the judgment of humans with the speed of machines is beginning to emerge on the Internet, challenging time-honored methods of searching that are wilting under the Web's explosive growth.
Traditional search engines have largely relied on using either the automated approach of indexing millions of Web pages, or the human approach of using editors to find and index what they deem are the best pages.
But as the Web has grown, both methods--the human strategy typified by Yahoo's directories and the automated approach used by search engines such as AltaVista and Northern Light--are too often frustrating exercises, sometimes returning incomplete results or thousands of useless ones.
Within the last few months, a hybrid strategy has surfaced with the development of search engines such as HotLinks, Simpli.com, Oingo and Ejemoni--each using a different method but with a common goal of drawing human intelligence deeper into the process of searching the Web.

In its own way, each of these projects is attempting to impose a greater sense of order and meaning on the vast anarchy that is the Internet.

"We're trying to get closer to having a computer think like a human," said Eytan Elbaz, director of business development for Oingo.

One of the biggest weaknesses of both the human and automated approaches is the reliance on keywords that search engines look for in their indexes. That's a problem because language can be so ambiguous. "Java," for example, can mean coffee, a programming language or an island.

Adding a few more keywords can help refine the search, but there is still the chance that some pages that deal with, say, employment--such as company job listings--would not even contain the word "employment."

Oingo (http://www.oingo.com), a Los Angeles-based company, and Simpli.com (http://www.simpli.com) of Providence, R.I., are among a small group that is trying to look at the meaning of the keyword itself rather than just the letters that make it up.

In addition to having an index of Web pages, Oingo and Simpli.com also construct a kind of dictionary that lists the meanings of various words. When a keyword is entered in a search box, a list of definitions pops up, drawing the user into the process to help narrow the search.

Because the dictionary also includes associations between related words, the engine can find relevant sites even if the keyword does not appear. By further refining the dictionary, the companies believe, they can give the computer a rudimentary sense of human language.

"We are looking at how the mind works and forms knowledge," said Simpli.com Chief Executive Jeff Stibel.

Oingo, which won a Best of Show award at the Fall Internet World exposition in New York this month, is already on the Web. Simpli.com hopes to start testing its search service by the end of November.

Ejemoni (http://www.ejemoni.com), a search company based in Redwood City, Calif., will begin a public test of its search engine at the beginning of next year.

This site also uses a dictionary, but it goes a step beyond Oingo and Simpli.com by allowing users to enter paragraphs or entire documents as search terms to give even more context to a search.

Chief Executive Nicolas Economou said entire documents can be scanned by the search engine, which evaluates the words in context and assigns the document a numerical value. The engine then searches through its index for a Web page or document that has a similar numerical value or "meaning."

"We are trying to understand what an entire document is about," he said. "The more context you give, the better it can zoom in."

Even with better keyword searches, the results can still be poor if a search engine's index of Web pages is not very good.

HotLinks, a Mountain View, Calif., start-up, has created a search engine that strikes directly at the problem, relying on the idea that what is important and valuable to people is usually what they mark for future reference.

HotLinks (http://www.hotlinks.com), which launched its search engine last month, does not actually search the Web. It searches the bookmarked links and favorites that people have saved in their browsers over the years.

The theory is that these links are usually the most important Web pages people have found in their journeys across the Internet. If someone else is searching for information, chances are good that these links could be important to that person too.

In essence, a bit of human judgment has been built into the HotLinks system. There is no complicated technology--just a lot of people opening their bookmarks to one another.

"The fundamental idea behind HotLinks is sharing knowledge," said Jonathan Abrams, HotLinks' founder and chief executive.

Visitors to the site can choose to send their bookmarks and favorite links to HotLinks' computer, where they are indexed so other people can search them.

In the next few months, the company will add a feature called HotLinks Guide, which will put together the most popular links on a subject in one category.

The company also plans to introduce a feature that will suggest Web sites based on where people have gone when searching on the same keywords--a process similar to Amazon.com's method of suggesting other books to look at.

"This isn't just a better search engine, but a way of discovering things without even searching at all," Abrams said.

Times staff writer Ashley Dunn can be reached at ashley.dunn@latimes.com.

Copyright (c) 1999 Times Mirror Company