Search engines cannot read minds (the plans on the white-board in Building 43 aside, of course). So, if we’re to type in “restaurant by the bank” I would get results about restaurants, financial banks, rivers and executive bonuses.
I would need to add search terms, and perhaps play that delightful search-engine guessing game we all know and love.
Improved ranking or relevance is one answer to this challenge, while it is no magic-bullet it certainly helps to have your search engine profiled to return results that match the kinds of things that users generally look for. Up-weighting certain documents formats, or pages with strong peer-ranking which include certain keywords is one way to work around this.
The problem is “generally”.
The Problem with Profiles
Profiles match a general group, and if they are applied too strongly, they make it harder to find anything that falls outside of that profile. Setting up multiple profiles that users can swop between is another approach, but requires still more mind-reading. A cheaper approach is to make the results easier to navigate by revealing as many of the facets actually present in the result set and then letting the user drill down through the results based on what is actually in the results.
Going back to my search for something eat near the bank, it would be much easier for me to see an overview of what sorts of results I got back immediately, rather than be confused about why I am seeing a whole slew of results about executive compensation (and expensive restaurant bills). While I might get results like that at the top of the list, I’d see the various facets in the results (banks, financial crisis, restaurants, railway-stations and rivers) and be able to zero in on what I am looking for.
Now, this obviously presupposes that the documents in your index actually have this data attached to them, but users are suddenly given a much wider overview of what sorts of results have been returned without having to click through page after page of results. Since this meta-data is returned based on results, it is not dependent on a “one-size-fits-all” profile. Though, of course, this is dependent on how the documents are tagged and categorized.
But, at least it’s not mind reading.