I'm trying to implement the naive Bayes classifier for sentiment analysis. I plan to use the TF-IDF weighting measure. I'm just a little stuck now. NB generally uses the word(feature) frequency to find the maximum likelihood. So how do I introduce the TF-IDF weighting measure in naive Bayes?
You can visit the following blog shows in detail how do you calculate TFIDF.
You use the TF-IDF weights as features/predictors in your statistical model. I suggest to use either gensim [1]or scikit-learn [2] to compute the weights, which you then pass to your Naive Bayes fitting procedure.
The scikit-learn 'working with text' tutorial [3] might also be of interest.
[1] http://scikit-learn.org/dev/modules/generated/sklearn.feature_extraction.text.TfidfTransformer.html
[2] http://radimrehurek.com/gensim/models/tfidfmodel.html
[3] http://scikit-learn.github.io/scikit-learn-tutorial/working_with_text_data.html
来源：https://stackoverflow.com/questions/6291546/how-to-implement-tf-idf-feature-weighting-with-naive-bayes