August 11, 2003 4:00 AM PDT

Searching for the personal touch

A stealth start-up out of Stanford University is hoping to raise the heat on one of the toughest problems in Web search--and possibly out-Google Google in the process.

Kaltix was formed in recent months by three members of Stanford's PageRank team--a research group created to advance the mathematical algorithm developed by Google co-founder and Stanford alum Larry Page that cemented Google's fame.

PageRank has helped steer people to Web sites like no other search technology before it, harnessing the link structure of the Web to determine the most popular pages. Now, Kaltix hopes to improve upon PageRank, with an attempt to speed up the underlying PageRank computations.

That, in turn, could lay the groundwork for a breakthrough in a cutting-edge area of Web search development known as "personalization," which aims to sort search results based on the specific needs and interests of individuals, instead of the consensus approach pioneered by Google.

"Kaltix is a 'stealth mode' start-up...(leveraging) research done at Stanford University as well as several new technologies developed at Kaltix to provide large-scale personalized and context-sensitive search," a Kaltix representative said, declining to comment further.

Kaltix has disclosed few specifics about its plans or technology. But the company's general statements appear to place it in a sweet spot for innovation that's being pursued by all of the major search providers. Now that Web search has become a moneymaker for portals such as Yahoo and Microsoft's MSN, technologists from all the industry players are back in the labs developing formulas to personalize search.

Web companies outside the search industry have long made attempts to create personalization features, but most of these attempts have fallen short of expectations. Amazon.com, for example, regularly serves up book titles related to a visitor's previous purchases, which may no longer be relevant. A personalization feature offered through TiVo, a maker of video recording devices, was criticized when reports circulated that the device would recommend gay-themed television programs to viewers based on just a few program selections.

Despite these flawed attempts, developers continue to have faith that personalization technology can be created that will ultimately unleash marketing and revenue opportunities.

If search developers are successful in building such technology, they could help millions of people better navigate the Web while raising the profile of numbers of obscure sites. Personalized searches could also unlock new revenue, from online advertisers seeking to maximize their return on marketing dollars.

The advertising-sponsored search business is expected to reach between $6 billion and $8 billion within five years. By creating a personal relationship with Web surfers and delivering spot-on results every time, search engines could improve response to links for advertisers' products and services.

"Personalization is one of the holy grails for search," Chris Sherman, editor of industry newsletter Search Day, said. "Everybody's working on personalization to some degree or another. When it comes out of the labs and what flavor it takes are the big questions."

This time, it's personal
The personalization of search tools entails matching results to user profiles. These profiles could include data such as zip code, birth date or individual search history. For example, the keyword "jaguar" might place car sites at the top of search results for someone who had recently visited automobile Web sites, but might lead off with Web sites about the cat for someone whose surfing history showed an interest in animals.

Personalization techniques include collecting data from the search user directly, as well as putting algorithms to work behind the scenes. With a little information voluntarily submitted by a searcher, an engine could localize search for results in German or French or segment listings to show only 15 out of the top 100 links. An advanced algorithmic technique might apply keyword-pattern analysis in order to examine an individual's search history and guess what the object might be of the next search request.

Many search engines already use some rudimentary personalization features.

AltaVista uses so-called geotracking technology to detect visitors' Internet protocol (IP) addresses and guess their geographical location. That can provide useful context for some searches, for example, in returning soccer-related results for a query on "football" from a user based in the United Kingdom.

Yahoo is also implementing personalized search features. Among other things, the Web portal has taken advantage of its relationship with visitors to deliver more tailored answers in specific areas. For example, its yellow pages, weather information and "My Yahoo" sections all use members' zip codes or other personal data to deliver tailored information.

"It's a key focus for Yahoo," company spokeswoman Diana Lee said. "Being able to bring a more personalized experience for visitors makes it better for them, for us and for advertisers."

Microsoft has publicized that it's working on advancing MSN and desktop search. One area of development could be in integrating and personalizing search across Microsoft Office, Microsoft Outlook and the Web.

Search leader Google has also shown an interest in the area. Two years ago, it bought Outride, a spinoff from Xerox's Palo Alto Research Center (PARC). Outride uses data-mining techniques, pattern recognition and natural-language semantic analysis to improve search results. But the acquisition has yet to produce visible results for Google.

Jim Pitkow, former CEO of Outride and current president of search company Moreover, said that personalization is an issue that Google is actively exploring. Google declined to comment for this story.

"The macro trend now is you've got the three main search contenders--Yahoo, MSN and Google--trying to make search better, and you can only do so much by just looking at the content and just looking at the links," he said. "In order to make significant breakthroughs that would be perceived by end users, these companies are all looking at personalization techniques."

Not so easy to build
If the potential rewards of personalized search are great, so are the hurdles to building a successful personalized search engine.

"The problem is, there isn't a 'one size fits all' formula," Search Day editor Sherman said. "By its nature, it's something that has to be tailored to each individual user. It's not like Google can build a personalization module, plug it in and flip a switch. It's a more-complicated effort, and it will require major investment and commitment. It's not clear yet that there's enough user demand to justify that cost."

The main task will be getting the user interface right. That means giving people notice of what data has been collected, where that data will be stored and how it will be used. It also means giving users the choice of changing data or removing it.

"A lot of it is going to be experimental. Personalization can turn people off if it's interfering and annoys them," said Amanda Spink, professor of information science at the University at Pittsburgh.

Though Kaltix has disclosed little about its technology, the start-up is attracting notice from search-engine veterans.

The company has demonstrated its service to veterans in the industry, including chief scientist Jan Pederson at AltaVista, which was recently acquired by Overture Services. Pederson said that Kaltix "was likely looking to get bought out."

Without discussing Kaltix's plans publicly, the company's founders have published research that claims to offer a way to compute search results nearly 1,000 times faster than what's possible using current methods.

Outride's Pitkow said Kaltix's method is similar to looking for a tree in a forest by examining only a clump of trees rather than the whole forest.

"If you can localize your computations to those types of trees then you can be much faster," Pitkow said. He added that it takes days to compute PageRank. "If you can compute it really fast, you can compute it on per-person basis," he surmised.

"If they've been able to take a computational block and remove it, that opens up new opportunities."

 

Join the conversation

Add your comment

The posting of advertisements, profanity, or personal attacks is prohibited. Click here to review our Terms of Use.

What's Hot

Discussions

Shared

RSS Feeds

Add headlines from CNET News to your homepage or feedreader.