Magazine article Information Today

Clusty Solves Information Overload

Magazine article Information Today

Clusty Solves Information Overload

Article excerpt

Google is now the 800-pound gorilla among search engines, but it won't necessarily continue to be this way forever. In recent issues, I've reviewed intriguing search engines, including Grokker (October 2005) and (May 2006), which have important functions that Google does not. This month, we'll look at Clusty (, an attractive alternative to Google and other prominent search engines, such as Microsoft Search and Yahoo!, that follow the Google model.

What is this great advantage that Grokker,, and Clusty all share? They are all--to use Clusty's term--"clustering engines." In addition to retrieving a set of Web pages that are relevant to your search query, they classify, or "cluster," them by subject to provide more targeted and focused results. They all use basically the same technique: They automatically scan the initial search set to identify recurring terms or phrases on the premise that subject-related pages use common language, and then place them together in separate folders. This premise is generally sound, and the execution is usually effective.

Clusty Makes Its Mark

Clusty began as an enterprise search product from Vivisimo (http://www, which was founded in 2000 by a group of computer scientists from Carnegie Mellon University. Clusty, a general Web search version, was released in September 2004. This year, Clusty has already made two major announcements: It was chosen to power, the government search portal, and it has offered to provide free site search service to nonprofits and educational institutions.

The first announcement, which was made in March, is big news for Clusty, because is a major Web portal to federal, state, and local government information. The free site search service has been implemented by several sites, including Carnegie Mellon's computer science department, which uses both Clusty and Google site search. Check it out at http://www and run your own side-by-side comparison test. Both the partnership and the site search service are high-profile initiatives may help Clusty move up in the ranks of the search engine race.

Of course, the publicity value of these announcements will be short-lived if the product itself doesn't live up to its hype. But Clusty is a strong performer on all of the measures of the modern search engine. It searches a large number of Web pages, its underlying search technology is effective, it has lots of bells and whistles, and the clustering option makes it the very latest thing, providing a compelling alternative to the search engine status quo.

Why You Want a Clustering Engine

Clustering search technology is not new. It was demonstrated in 1996 on the prominent Web search engine Northern Light. Since then, Google's undeniable utility has drawn attention away from everything else.

Google does one thing supremely well: It produces a very useful ranked list of Web pages. Although this serves one information need to quickly get a few highly relevant pages out of the chaotic mass of the Internet--it doesn't handle another common and important need--to explore the different aspects of a complex subject. This is where clustering engines come into play.

Clustering engines actually provide the best of both worlds: First, they create a single ranked set of search results for when you only want a few high-relevance pages from the top of the list; second, they create a hierarchical subject classification that enables you to go directly to the pages dealing with a particular aspect, instead of working through a long, undifferentiated list of links.

But clustering engines have their own pros and cons. They are particularly good at breaking out subjects that have broadly differing aspects; they are less able in making smaller distinctions because machine-sorting technologies are clumsy when dealing with language's finer nuances. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.