With the development of SVM，its application has been extended from binary class classification problem to multi-class classification problems. Its concepts and algorithms become various and different. The main study in this paper is DDAG, which is one of the popular algorithms, whose principle is combining the binary class classifiers with multi-class classifiers. This paper also points out the strengths and weaknesses of DDAG, when it is used in multi-class SVM text classification, and presents an improved algorithmic concept to make up for the weaknesses.