I've been working lately with SQL Server Denali's Semantic Search feature. There's a more complete description of this feature (as well as how to set it up, its an addition to full-text indexing) in BOL, but I just wanted to summarize my one sentence "raison d'etre" for it on a function by function basis. And mention a couple of the more obvious use cases.
First, the three functions:
SEMANTICKEYPHRASETABLE – Without a source_key argument, this is "show me all the keyphrases that appear in my entire corpus of documents". With a source_key, just substitute "…in this particular document". You have to know what's in your document collection, after all.
SEMANTICSIMILARITYTABLE – Given an exemplar document, this one is "show me the (usually TOP N) documents that have similar content to this one" on a scale of 0 to 1.0.
SEMANTICSIMILARITYDETAILSTABLE – Given two examplar documents, this one is "show me why you think these are the same" in terms of specific semantic keyphrases.
Here's a use case. Want to hire the perfect job candidate? Make up an exemplar resume in a standard format, filling in the qualifications of your perfect candidate. Then find, in your existing resume pile, the ones that are most like your perfect candidate. Check your work by checking the matching keyphrases for each of the TOP N candidates. Be careful though, candidates have ways of knowing what keyphrases you might want to hear… Do the interview anyway. And keep track of how semantic search's predication was.
Here's a couple more. Reading a book? Buying a product? The library or the store doesn't have *exactly* what you want? Find the closest match. And then the answer to why its the closest match.
And finally, a friend of mine, Andrew Fryer, did a blog post and set of videos about semantic search. Sounds like (from the name of the table in his blog post) that he might be matching up similar PowerPoint decks. Or at least the table is called MyDecks.
Hope this gets you going with semantic search. Can't wait till they can find similar songs and pictures, a la their presentation at PDC about it a few years ago. Maybe next release?