Next frontier: searching images

So much remains to be done with search technology. But in a world dominated by Google (and to a lesser extent Yahoo, Microsoft and Ask.com), a young startup with a hot idea shouldn't challenge the big guys head-to-head.

A more realistic approach is to develop a service that would introduce new search methods, while understanding that in today's market, good ideas that get noticed tend to get acquired. YouTube's founders learned how lucrative that can be.

Image search is a case in point.

Sure, you can search for images on the Web, but the search relies on keywords that are associated with the image. True image search is tough, because computers don't know how to analyze the objects they see, making a breakthrough in this area something of an obsession with search experts.

At a recent conference, Google officials discussed what they call "VisualRank," a way of combining improved image recognition with the kind of analysis that Google runs on Web sites. This produces images more relevant to a search.

But all this is at the research stage, an indication that even with Google's formidable resources, we're in genuinely tricky territory.

Better, perhaps, for an aspiring search company to use images in a different way, by re-thinking how to present search results.

Slick Viewzi

This is what Viewzi of Dallas (viewzi.com) does. It offers of a list of results, a set of "views" based upon basic categories of information. Run a search and you can choose among different ways of drilling down into the information via an attractive interface.

My search, for example, might demand news items, in which case I can choose, from the Viewzi results, a view tapping Reuters.

The design is slick. Thumbnails help you determine at a glance -- without having to enter a site first -- whether the site will help you. Views include photos, videos, text, combined searches (tapping multiple engines), shopping and weather.

Viewzi is still in beta testing, but the demo shows its capabilities, and you can sign up to gain access to the full site.

How we package search results is part of the picture. How we arrive at them is the other.

The two come together interestingly in Powerset (powerset.com), the much-anticipated search company that is trying to use semantic methods to catalog, and in a sense "understand," the content of Web pages.

Powerset impresses

Powerset has rolled out its first engine using these methods. It's hardly a Google-killer.

In fact, the current design does nothing more than search Wikipedia and the open database Freebase, which on its face seems unimpressive, because Wikipedia has a perfectly good search function of its own.

But play around with the site a bit, and prepare to be impressed. Enter a keyword, and you'll get a results page that includes semantic pointers (Powerset calls them "factz") about your subject, each built on relationships that the engine has discerned about the topic.

You can quickly click into specific information, no matter which Wikipedia article it appears in, a process that saves time over conventional search methods and is more accurate. You also get useful article outlines amid results that can seem eerily targeted, because Powerset doesn't just follow keywords but the concepts behind them.

Entering a specific search question ("What cities did the Doolittle raiders bomb?") often produces lightning-fast results.

So what is the strategy behind using Wikipedia to promote what could be a Web-wide search engine?

Clearly, Powerset is making a virtue of necessity, the necessity being that its indexing methods are hugely intensive. Remember, it's trying to figure out what pages are about so that it can answer a query, even if the page it finds does not contain the specific keywords entered. Showing how the technology works on a limited basis keeps the company relevant to search users as further financing helps it expand.

I would imagine Powerset is going to be attractive as an acquisition, based on its current work alone. There are rumors about Microsoft's interest.

But I also wonder whether this model will be ideal for targeted search in the corporate environment, where querying internal information sources could be made far more efficient. Play around with this engine, ask it questions, and you get the feel of an effective new interface.

Many of these ideas will develop more slowly than we expect, given the magnitude of the task -- 20 billion Web pages is no small thing to index when you're mining more than keywords.

UpTake for travel

We'll see companies trying out new methods in specific markets such as travel. Palo Alto, Calif., startup UpTake (uptake.com) hopes to move beyond travel booking sites to extract concepts from a vast database gleaned from millions of reviews and descriptions of hotels and other travel services.

Like Powerset, UpTake uses semantic analysis to take you where you want to go. It's clear that creating connections between data by indexing concepts and meaning will have ramifications across the search spectrum.

Yahoo, Microsoft's once and perhaps future target, has been actively working with enhancing its search results, but I'm wondering whether the methods Powerset can bring to the table won't ultimately prove more useful to the company. It's surely more of a gamble than Yahoo but acquiring it would be a reasonable move, perhaps what Microsoft needs to face up to the Google juggernaut.