Can I use only idf and skew value of idf to predict the importance( trend /probability ) a function in a large no of text data instead using bm 25 scoring. I do not need a group of lexical keywords matching for my purpose. Can I get a recent ref paper supporting only idf rank?