They way I see "semantical relevance" I can only compare it to graphs where words are the "the nodes" and a "path" that conect two or more of those words. Then those paths are associeted with a value (for relevance between two specific nodes).
With enough resources available, following that "semantical" ghrap with an algorithm such as Diskjra, will find the shortest rout to "relevance"....
May be I'm just in my own little world with this theory, but if there's any resemblance to the "real" way of weighing
semantical relation-ships... we may as well forget to "study G" and start "styding our users" very closely.