Email	Print Reprint
	add to:
	Del.icio.us Digg Google Furl	Slashdot Y! MyWeb Blink

Speaking My Language

Natural-language search sites have created their share of buzz. The key component of InQuira's newest product, InQuira 5 for Search (www.inquira.com), is the ability to interact with users in a natural, conversational way. Similarly, iPhrase understands that users who ask questions like "What is yield?" are looking for an explanation; the first result says, "Here is a definition for yield." If it's clear what the user wants, the search engine takes users straight to a page where they can take action with the information instead of presenting a page of search results. For example, say a user types in "What are Cisco's projected earnings?" A page will be displayed that contains a chart with the numbers. Pino claims a customized, fine-tuned iPhrase package gets 95 percent first-page relevancy, compared to 40 to 60 percent for typical search vendors.

Part of the battle is getting users to feel comfortable writing queries that are full sentences. "It's the crevasse between the user and the search engine that's the problem," says Sue Feldman, an IDC analyst who specializes in natural-language searches. "Queries are often very poor sources of information for the search engine because they're typically one to two words."

But good results clearly make a difference. Schwab, iPhrase's first customer, noticed a change in the behavior of its 7.5 million end users within a few months of installation in June 2001. "When they went live, only about 25 percent of the queries were more than one word because users were conditioned to do that with traditional search engines," Pino says. Within several months, the number had risen to 75 percent, because customers were getting good results on their searches. At the same time, customer-service calls dropped 13 percent, saving Schwab an estimated $1.2 million.

Many other players are tinkering with similar concepts. Forrester's Sonderegger says he is impressed by the way Endeca and Verity (www.verity.com) now handle structured content, and by Recommind's (www.recommind.com) probabilistic approach to handling unstructured data. Teoma (www.teoma.com) uses subject-specific popularity, a trademarked technology that ranks a site based on the number of same-subject pages that reference it. Inktomi (www.inktomi.com) recently acquired Quiver and incorporated its categorization and taxonomy technology improvements into a new enterprise search solution. And in June, Jeeves Solutions (www.jeevessolutions.com) introduced JeevesOne Enterprise, a search product that directly connects structured information sources, like CRM and legacy applications, to a user's query.

Seeing the Light

Images present one of the most vexing search conundrums. Go to Google Images and type in "rose," and you get about 197,000 hits—with everything ranging from the flower to people named Rose to a Rose Garden seating chart. "If you see one or two that you like, and you want to see something similar, there's no way to tell the search engine to do that," says Michael Crandall, vice president of business development for VIMA Technology (www.vimatech.com).

VIMA's founders (who recently changed the company name from Morpho Software) faced down this daunting task by developing software that recognizes features like color, contrast, and texture, and also learns from user feedback to hone the search. The result is what VIMA calls "perception-based image recognition."

As with Google, users are first presented with a series of thumbnail images. But here's where it's different: Users can select one image and ask the search engine to find similar ones. The software examines 150 features in the image and searches for matches. "What the software is trying to do is reduce the image to some sort of fingerprint using a fast algorithm," Crandall says.

The technology can also be used in reverse: searching for images that users wouldn't want, and stopping them from appearing on a site. VIMA developers have fed the software ten thousand pornographic images that it can now recognize and block. "If you've got a brand name with an established auction site, for example, you don't want your brand sullied by any number of things," Crandall says. It can also filter, for companies, what their employees can see on the Internet or in email.

He acknowledges that such uses could spark a freedom-of-speech debate, but points out that the tool can bolster sexual harassment policies and related issues. In October, Taiwan-based BroadWeb Corporation deployed VIMA's visual search engine in its high speed Internet content filtering product called XKeeper, which gives high throughput (2GB per second) performance for one price, regardless of the number of users.

Previous Page | 1 | 2 | 3 Next Page

RELATED ARTICLES Nozzle: Counteracting Memory Exploits SDK, .NET Controls Released for Communications Enabling Web Pages Welcome To the World of Cryptographic Voting Jetbrains Updates Ruby IDE Altova Updates XML Editor		TOP 5 ARTICLES No Top Articles.