TF-IDF vs. Bag of words (BOW)

Question

I have trained my model with SVM and Multinomial Naive Baye. For both classifiers, I have used the TF-IDF and BoW feature extraction. The result of TF-IDF is better that BoW