Selecting query terms that best distinguish relevant documents from irrelevant ones in a specific corpus.