I think you meant TFIDF, but that would only work if you were looking for salient phrases in an article from many articles. When you look at a corpus of news the algorithm would likely work less effectively than an algorithm that using linguistics, for example Topia or even Rake.
