You are given an online review dataset with a binary classification task of positive vs negative reviews. You have to perform text classification using Nave Bayes model. You perform cleaning and preprocessing on the review data. Then using bag of words you prepare the data to feed to Nave Bayes model. You have seen no training documents with the word fantastic and classified in the topic positive. So P("fantastic"|positive) = 0. Because of conditional independence this zero probability cannot be conditioned away. How do you fix this problem so that your posterior probability does not become zero?