Relevance has always been the main goal of search for most of us searchers, although sometimes completeness can be even more important, e.g., when we want to determine relevance ourselves and volume is not an issue. Relevance is relative, and there is no way to write code that can anticipate relevance in a general way. (Of course quality is relative too!) Fortunately, search can be extremely useful even without the mind reading option – in fact, mind-reading wouldn’t be enough to anticipate relevance enough anyway.
Much of the discussion about search quality these days revolves around the front-end of relevance, i.e., determining, as much as possible, searchers’ intent. And we do have increasing amounts of information (such as surfing behavior) that allows us to make better guesses about intentions.
We can also make information richer so that search engines can make more accurate determinations about relevance. For example XML site maps provide context in the form of structural information; providing additional metadata to search engines provides even more context.
Despite the imprecise, and constantly changing meaning and use of language, we have been able to asymptotically improve our ability to determine both intent and relevance, and incrementally improve search quality.
I say “we”, but I am neither a developer nor an expert on search technology. We are fortunate to have someone who is arguably the most influential expert and developer today speaking about search quality in two weeks at our San Francisco conference. Udi Manber, VP Engineering, Search Quality, Google is going to open the conference with a presentation on Search Quality and Continuous Innovation. While Udi won’t be giving away any secrets, his presentation will provide valuable and fascinating insight into the way Google thinks about improving search quality. For a taste of Udi’s clear and straightforward style, and what he’ll be talking about read his recent blog post: Introduction to Google Search Quality
