February 26, 2001 11:20 AM PST
Search project prepares to challenge Google
Web search technology has long been vexed by the problem of prioritizing the thousands of Web pages that common queries yield. Project Pandango, the work-in-progress of Seattle-based i5 Digital, attempts to determine that priority by examining the Web surfing histories of a wide network of computers on the Web.
The ambition driving Pandango's creators is to take the peer-to-peer model, made famous by file-swapping applications such as Napster, and build a search technology on it that will topple the current market leaders.
"This is the third generation of Web searching, and Google has to be concerned with how they stay ahead of the curve," said Liad Meidar, chief executive of i5 Digital. "Being able to leverage a P2P network to do a search is taking what Google's done to the next level."
i5 Digital is putting the finishing touches on Pandango while pursuing negotiations with what Meidar describes as "a lot of the household names in search portals, ISPs if you will, Web-based companies with large numbers of users," Meidar said.
Because Google powers the search feature on Yahoo, the Web's most-visited site, it is considered the reigning king of search technology. The company also runs its own Web site.
Meidar is quick to point out how Pandango's approach differs from those of its potential competitors. Google, for example, is based on traditional academic citation analysis, which determines the relevance of a scholarly article based on how many times other articles cite it. Google ranks pages based on how many other pages link to them.
Pandango, by contrast, would determine relevance by examining a radiating network of "referrers." Once someone downloaded Pandango and joined its peer-to-peer network, that person's keyword search would examine the Web histories and bookmarks of an initial network of 100 referrers. From there, the application would search the Web histories of those 100 referrers' combined 10,000 referrers, and once again--so that the query would canvass the Web pages visited and bookmarked by 1 million people.
The system is being designed to refine itself over time based on which pages--and by extension which referrers--the searcher selected. Frequently selected referrers would move to the top of an individual's referrer heap, while others would drop off and new recruits would join. In the process, the Pandango member would be developing an initial referrer group of like-minded Web surfers, whose once- and twice-removed referrers would be increasingly like-minded as well.
A P2P collaboration
If Pandango is a child of the peer-to-peer craze, then its other parent is the more tried-and-true Web marketing technology known as collaborative filtering. Familiar to shoppers at e-commerce sites such as Amazon.com, collaborative filtering analyzes people's behavior to speculate on preferences and make recommendations.
In the Amazon example, the technology lets the company look at what one customer has bought and recommend books based on what other buyers of that book have purchased in addition to it.
"This was about finding a way to find a group with common interests," said Erik Fretheim, chief technical officer for i5 Digital. "The idea comes from collaborative filtering, the concept that people who are like you, who have similar interests, are probably the best guides to what to find."
Fretheim likened the Pandango search experience to common word-of-mouth queries.
"When you come to a new town looking for a doctor, you could look in the phone book, which is like your typical search engine. What you do instead is you ask your peers, people you know, for recommendations rather than just take a blind shot at something."
Although Pandango would likely sidestep the legal quagmire that has tripped up Napster--and frightened other peer-to-peer ventures--analysts say i5 Digital will have to be creative in coming up with ways to make money with the project.
"I really like the technology, but whether or not it's a business model is the question I have," said Scott Pope, a Boston-based independent analyst and consultant until recently with the Delphi Group. "It's an interesting concept, but who's going to pay for this? It may wind up being more appealing to the people in academia or to consumers rather than major companies."
i5 Digital hopes to license the Pandango technology to companies that would pay to use the application to facilitate communication among employees, who might eliminate redundant efforts if they could see what resources their colleagues have mined.
The start-up is also eyeing the licensing market that Google enjoys, reaping fees from portal giants including Yahoo. i5 Digital is also considering an outright sale of the technology.
The membership search
But although Google's ascent from obscurity was swift and recent, catching up to it will require some time and widespread adoption, Pope warned.
"There's a huge inertia that needs to be overcome," said Pope. "To make this technology work for the general public, you need millions of people to use it and install it, which is no easy task, quite frankly. The best opportunity for their product is with knowledge workers in highly paid positions who do a lot of information gathering. That's where their biggest value proposition comes into play."
Not surprisingly, potential rival Google weighed in with similar reservations about peer-to-peer's search potential.
"I have to agree that P2P is an interesting technology, and Google is looking at it," said Craig Silverstein, director of technology at Google. "But there are serious problems. For some applications this works great--if it demands resources that people don't care about giving up. But for search, it's using resources people care about very much. Going through documents takes up memory. People are very loath to use something if it's going to interfere with their word processing."
Indeed, like any peer-to-peer application, Pandango would take as well as give. People who downloaded the application would be expected to pull their own weight and assent to becoming a referrer whose Web history would be searched in network queries.
But Pandango promises a system that is both nondemanding of individual's computing resources and incapable of violating anyone's privacy.
"It's much less intrusive than most P2P applications," said i5 Digital's Meidar. "Pandango does not go through documents for Web searching--so none of the users desktop applications are used in the process. We don't look through people's hard drives like Gnutella or Napster. It's just a limited amount of information about their Web behavior. On top of that, it's fully anonymous for both requests and results. We're a fully distributed system, so it's not going through any type of server. We don't work for the FBI."
Another problem facing the developer of a peer-to-peer application is that by its very nature, the finished product lies somewhat out of the creator's control. As a result, i5 Digital is keeping Pandango on a tight leash as it negotiates with potential partners.
"It's a Gnutella-type project," he added. "It's open to the public. Once it's out there, it's not intended to have any centralized server you could shut down. It's set up so that we don't necessarily lose control of it, but once it's out there people can try to pull it apart and mimic what we're doing."
A branded version of the product would let a company exert some control over versions and upgrades, Meidar said. In addition, the company has a patent pending on the technology.
Giving technology a second chance
Pandango is one of several projects in varying stages of development at i5 Digital. Although Pandango was developed by in-house engineers, the company's specialty is reviving cutting-room floor technologies, or those that were created by a research and development effort but never wound up making their way to a market.
Billing itself as an "intellectual property discovery and development company," i5 Digital approaches universities, large corporations, government agencies and other research centers with the history of Xerox Parc as a cautionary tale.
"The classic example is Xerox Parc," said i5 Digital's Fretheim. "They created the Ethernet and windows, but they were never able to commercialize them. They didn't try to--they just said 'these are things we don't need.' There's a lot of that technology out there, and we look for that kind of stuff."
Unlike Xerox Parc, which failed to profit from many of its famous inventions, research groups working with i5 Digital get a cut of future profits, either with an equity stake in companies founded around their technologies or through royalty payments.
What i5 Digital describes as its "most mature" project is DealPlanner, an application service provider for organizing large transactions such as mergers and acquisitions, initial public offerings, and large venture capital deals. DealPlanner would let investment bankers, lawyers and executives collaborate on document sharing, due diligence, working group management, scheduling and other tasks.
The project has a few "very high-profile" partners who will test its service, and a beta version 2.1 of the system will go out to beta partners next month. DealPlanner has a patent pending.
Project Looking Glass is working on technology that would give outdoor media companies a better idea of how many people are looking at their advertisements. Based on the work of engineers who were refining traffic safety and management technology--such as mechanisms that control the timing of traffic lights--Project Looking Glass aims to analyze the audience of outdoor media in finer detail than is now possible, including their demographics.
Project Brimstone is a firewall application based on technology developed by i5 Digital for other purposes.
Project Mars is turning technology developed at NASA for its mission to Mars into an application for use with wireless communications infrastructure.
"The particular researcher we were talking to said the technology would never have any application on Earth," recalled Meidar. Although the company hopes to prove him wrong, Meidar conceded that Project Mars is not in active development.
In addition to its Seattle office, i5 Digital keeps an office in Tel Aviv that works with universities, start-ups and defense institutions there. The company's research partners include NASA, the universities of Washington and Minnesota, and Technion in Haifa.
Whatever the success of i5 Digital's products in the marketplace, the company's mission has won plaudits from researchers with whom is has collaborated.
"It provides a way to commercialization for university research which otherwise would have been shelved due to university bureaucracy," Nikos Papanikolopoulos, associate professor of computer science and engineering at the University of Minnesota, said in an e-mail interview. "I think that the potential for impact is very significant. It provides a valuable incentive and gives to university researchers the real-world feedback that they often need."
The benefits of dealing with a company like i5 Digital aren't without their side effects, however.
"One potential negative impact--there are some ways that this can be averted--may be an increase in the commercial influence in an academic setting," warned Papanikolopoulos.