I am doing some information extraction study on some data. We extract a "word" list from the data. And for each "word", it has a n dimensional feature vector in R ^n. Next, We want to rank the list according to a score. The question is, how can I combine all the feature values to build a score function for my ranking. The difficulty is we can only work in an unsupervised way. I think many people are facing to this question in bioinformatic or other area when doing IE, right? I had thought about max entropy, but it usually works in supervised way. Is there any good method to solve this problem in unsupervised way ?