DBpedia recently added a search facility that beautifully shows off the power of structured searching--treating content on the web as a database, rather than as a big pile of documents.

Google-like, you have a simple, one-input interface to start your search over the DBpedia set of data, Wikipedia, put into the nice structure of RDF (i.e., semantic web). Un-Google-like, the results pull up an image for the hits if one exists (a very nice touch!). And, even better, you get a tag cloud at the top that you can use to filter the results. So, for example, type in "education" and you get a tag cloud that includes such filters as "academic degree," "professional association, "civil right," "government department," and many more. Click the filters to narrow down your search.

As is to be expected, there are plenty of unexpected tags in the cloud. The tag "party" in the same search, for example (it leads to "Liberia Education and Development Party"). Or "album," leading to a number of albums with the word "education" in the title. That will be both a strength and weakness, depending on the user. To people with a strong training in traditional library searches (lots of booleans and classification systems), it could be a little disconcerting. But, if you put that aside, it provides a way into the data that is more intuitive in some ways. Say, for example, I heard a really great song, but the only thing I remember about it is one word from the title of the album. . .

It also exposed a searching methodology I hadn't anticipated. One-word searches, followed by filtering with the tag cloud, yielded better results than trying to put together a string of keywords that I thought would get me there. Google (and for that matter most every non-traditional search) has me trained to search by piling on as many possible keywords as I can. Here, the filtering through structured web techniques makes that the wrong approach. Instead, one word to get me in the right neighborhood is the best plan, then the tags will guide me through what's actually there, rather than my best guesses about what I hope is there.

It seems to me that other structured web applications could allow the same principle to work, based perhaps on multiple clouds, where each cloud is a slice-and-dice around a different axis. So, in addition to the "tag" cloud, maybe a "foaf:created" cloud, a "sioc:has_member" cloud, a "skos:broader" cloud, etc. That would be more complex than the general user might want, but for more particular audiences it would work well, I think. Or, make that the next generation of "Advanced Search" options.