Query refinement generation via neural nets

Scoble laments that when asked “new york hotels” search engines do not know the difference between “hotels named new york ” and “hotels in new york ” (or, to some extent, “new hotels in york ”). Scoble wants the initial search-engine return to include questions intended to refine the search. Danny Sullivan agrees, saying that Ask Jeeves had such a thing, but the cost of humans creating relevant questions was difficult.

 

Here’s the solution, I think. You just throw a bunch of question templates “Are you looking for the history of X?” “Are you looking for reviews of X?” etc., hook them up with random Bayesian connections, and update them as necessary.