What are Hybrid classifiers used for sentiment analysis? How are they built? Please suggest good tutorial/book/link for reference. Also how are they different from other classifiers like SVM and Naive Bayes?
In sentiment analysis you may want to combine a number of classifiers. Let's say: a separate classifier for emoticons, another one for emotionally loaded terms, another one for some special linguistic patterns and - let's say - yet another one to detect and filter out spam messages. It's all up to you.
You can use either SVM, Naive Bayes, or anything else that best suits your problem. You may use majority voting, weights (for example based on cross validation results), or any other more advanced technique to decide which class is the most appropriate one.
Also, googling for hybrid sentiment returns tons of papers containing answers to the questions that you have stated. Please, don't ask us to rewrite this papers here.