About 8-10 years ago I read about Latent Semantic Indexing, which basically, in a nutshell, is a technique that compiles the entire World Wide Web in text format and, when you search for a certain combination of keywords, draws a relevance "vector" into the Web. Some of you may remember that Google used to display percentages next to each search result (99%, 98%, etc...) based on how relevant the result is to your search. Those percentages were supposedly based on LSI vector analysis.
This technique has been modified and improved over the years but, to my knowledge, it still exists today and is, in one way or another, used by all major search engines.