The emergence of the cloud computing has resolved the difficult of storing the abundant data and analysing data processing effectively. Based on the Hadoop open-source implementation, the cloud computing clusters distributable systems. Meanwhile, the usage of MapReduce parallel programming model has implemented a modified distribution on TFIDF Naive Bayes text classification algorithm. The experimental results show that improved TFIDF has chosen this unique method. The Distributed Hadoop framework has based on Bayes text which classifies automatically. This new achievement can not only handle the failure of nodes, but also possess high reliability and much more scalable advantages.