An automatic relevance and weighting method in which terms in a text or other content object are modeled as random variables, so that term frequency and distribution are used to predict the probability of relevance.