Abstract: Hierarchical classification is an effective method to solving the classification problem on the large-scale text data and it can save time without reducing the classification accuracy. A relaxed strategy combining the traditional hierarchical classification method to improving the system performance was introduced. During the process of hierarchy structure construction, the node juedgement of the uncertain category was delayed until it was classified clearly. The "block" problem was effectively alleviated to transfer the classification error from the higher level to the lower level in the hierarchy structure. The experimental results show that the relaxation approach can build a more reasonable hierarchy and further improve the classification performance. Compared to the other classification method, such as Support Vector Machines, the method has more advantage in time performance and is more efficient for large-scale text classification task.