Beyond Google: Collective Searching

Continued from page 1

The first time users visit a Boxxet, each link on the results page will include a chance to rate the content, starting with the question "Is this relevant to [your topic search]?" The user can then click "Yes," "No," or "Skip."

Every link includes a Like/Dislike rating button and a series of possible reasons why. For instance, you might rate something negatively because it is irrelevant, out of date, spam -- or just because you didn't care for it. The application scores each rating and adapts its searches accordingly. (If there are no results, the site will soon offer people the ability to create their own Boxxets.)

The Boxxet engine, which is built around standard software, such as MySQL and JavaScript, is driven by "thousands" of variables, according to Tsang, which he and his co-founder developed. The primary technology they used for categorizing searches is "support vectors," which break text apart to analyze its patterns, Tsang says.

In the future, Boxxet will also learn how to make best-of lists. For instance, if you rated a list of 30 movies, it would return recommendations based on ratings by others who've rated at least some of the same movies. Or the feature could be used to parse news sites and bring back the best stories and blogs on a topic.

On a practical level, Boxxet is more like a search engine than a social network. It's a concept that might appeal to people who're tired of trying to figure out how to refine searches on sites like Google. But it could also have some stiff competition from Microsoft, which just announced MSN Macros, a collection of search engines geared toward specific subjects.

"You can now author your own search engine and put it on your website," says Christopher Payne, Microsoft's corporate vice president of MSN Search. The company has introduced 43 "search macros" this week, and it plans to introduce many more over the coming months, in addition to those created by users.

For that matter, Technorati is also aggregating more types of content, notes Clay Shirky, an assistant teacher of telecommunications at New York University and long-time observer of social software. He says aggregation engines like Boxxet are needed on the Web. "Right now, search means you pull things out of context, which can demean their value. Anybody who can figure out how to aggregate things without losing context will have a very high-leverage application."

For his part, while Tsang thinks Boxxet's ability to let users drive what it does will make it the right product, he acknowledges that "we have to prove that people will come, that there's something here."