I have a document corpus and I want to estimate the probability of occurrence of a certain word $w$. Simply calculating the frequencies and use such a number as an estimation is not a good choice. Is there any work on this topic describing a better approach?