A quote from a recent article (1):
“...When you search for a crow on Google, you find plenty of references to people named Crow and people who crow about things, and that doesn’t help when you’re studying the bird,” Dr. Parr said..."
What should we expect? A mind reading Google that knows one is looking for Crow the bird, and not Crow the actor?! Results should include both to cover all bases.
It is the lack of context, or rather, the lack of opportunities to provide context during the search flow that leads to poor results. Moreover, even the more sophisticated approach that facilitates narrowing results through faceted browsing does not scale when the categories of knowledge are inclusive, overlapping or cross-disciplined.
Search is easy when both user and search agent know the answer as well as the question BEFORE the search executes. By that I mean the the user KNOWS what the answer should be. This is like searching for your car keys: You know what it is, you don't know where it is. When you find some keys, you know which is a car key.
When the search object is less defined and the search engine confused due to ambiguity, there is, naturally, a lower probability for a satisfactory result yield. Similar to grabbing the key chain a building engeneers carries, things are more complext because there are gazillion keys and they all more or less look the same. Which is the right key?
While a lot of thought is given to providing better interfaces for narrowing and navigating results, little seems to be on on the interface of search input -- the initial oppertunity to provide context.
(1) 'Helping Computers Search With Nuance, Like Us' , NYT, Peter Wayner, Sept 12, 2007