Searching Beyond Search Results

A new service mines the contents of Web pages looking for meaning and relevance.

Search is one of the main tasks performed online, and yet often it doesn't work as well as it should. Take, for example, the common experience of "back-clicking," when a user has to return to the results page several times before finding the information she's looking for. According to a 2009 comScore survey, 30 percent of searches are abandoned in frustration, and two-thirds of the rest required users to refine their queries before getting the desired result.

Smart search: Yolink attempts to identify relevant information on Web pages. This example shows terms picked out on Craigslist. Credit: TigerLogic Corporation

A new product called Yolink, which launched this week, aims to help users figure out which search results are most relevant. It does this by looking at the contents of the Web pages that a list of search result link to. The company bills itself as a step toward semantic search, because it attempts to find meaning in the contents of a Web page. And it can do this even though most pages aren't marked up in the formats typically used to help machines interpret content. The product is made by TigerLogic, a company based in Irvine, CA.

Yolink's technology is designed to look past the links on a search results page and perform the next few steps in assessing the value of information for the user. Brian Cheek, vice president of business development for TigerLogic, explains that the company has built several demonstrations of how its technology can be used to enhance the search functionality on a site. These include one for Craigslist that pulls out data from apartment rental listings such as whether a unit allows pets or has parking, using this to supplement the results that the site already returns.

Story continues below

Jeff Dexter, who is director of product development and technology chief architect at Yolink, explains that semantic search typically calls for the text on a page to be marked up by the publisher with tags that help identify different types of information. Yolink is designed to deal with data that hasn't been marked up. Dexter says the company tried to establish a common model for how data is typically presented across a website. That model is used to identify particular pieces of data.

Yolink will analyze a page--such as a Craigslist listing--looking for clues as to where certain pieces of content are shown. It might notice, for example, that location and price information tend to be near each other. Search engines such as Google present snippets of pages beneath the links they provide to perform a similar service for the user, but Dexter says Yolink's results go beyond simply identifying keywords and surfacing them.