Content-aware searching

When I'm deeply engrossed in R&D, as I have been lately, I can become obsessed. So take what I’m about to say with a grain of salt, but I really think I’m on to something — namely, content-aware search.

Way back in the last century, during XML’s formative years, an oft-heard argument for XML was that it would enable smarter Web searching. Well, it didn’t. One reason we don’t have Web searching that exploits structured content is that we never got ubiquitous and easy-to-use writing tools to create well-formed XML content. So there aren’t many pools that can be plumbed with XML-aware search technology.

Another reason is that Google is good enough. At InfoWorld’s 2002 CTO Forum, Google co-founder Sergey Brin threw cold water on the idea of instrumenting content for intelligent search. "I’d rather make progress by having computers understand what humans write," he said, "than by forcing humans to write in ways that computers can understand."

Brin’s pragmatic stance sharply opposes the idealistic view of the Web’s inventor, Tim Berners-Lee, who continues to evangelize his vision of a Semantic Web full of carefully encoded content that we can precisely search and fluidly recombine. My own humble contribution to this debate is a prototype search engine, now running on my Weblog, that tries to steer a middle course between the Scylla of simple fulltext search and the Charybdis of unwieldy tagging schemes and brittle ontologies. The twin enablers of this prototype are XHTML content and XPath search. Because I maintain my Weblog’s content as XHTML — HTML’s well-formed cousin — I can query it using XPath patterns. That means I can answer questions that I can’t answer with ordinary full-text search. Some examples of things I can find this way: paragraphs with links to book-related Web sites, tables with more than five rows of data, and articles with references to audio or video clips.

These queries don’t depend on any special HTML coding. They require only that the HTML be well-formed XHTML. Of course, the vast majority of published HTML isn’t well-formed. Does that make this approach a non-starter for most repositories of Web content?

Not necessarily. The next phase of my experiment involves converting the 200 or so Weblogs I scan everyday, using my RSS feedreader, from HTML to XHTML. Early indications are that this will work reasonably well.

Remember, the pools of HTML content that your people routinely create, and the infinitely vaster pools to which they have access, are full of intrinsic metadata — including the links, tables, images, and other elements that occur naturally within HTML content. Mining that metadata may be more practical than you think.

By implementing content-aware search against existing repositories, you can show people the tangible benefits of more expressive content. Months ago I began writing Weblog entries that identify the sources of quotations, and the programming languages in which included code snippets are written. This was a promise to the future. I knew that I’d later be able to find these things very precisely, and now I can. But most people live in the present. For any extra effort, however modest, they quite rightly expect an immediate payback. Content-aware search is a great way to reward such effort.