Data analysis of the textual data floating through social networking sites is being viewed as a promising source of knowledge people - the potential users and consumers of certain web application. The existing text mining techniques need to be correct and fast. Also, statistical classifiers like Naive Bayes are fast and easy to implement but do not perform well for imbalance text datasets or for datasets with highly correlated features. In this paper, a modified model for Naive Bayes classifier for multinomial text classification has been proposed by modifying the conventional bag of words model. The experimental results over benchmark datasets prove its superior performance than original Naive Bayes multinomial model. Feature selection and term weighting is combined with the proposed classifier for studying how well it can be implemented for various text mining applications.