Hierarchical multi-label classification is a type of classification which combines two aspects of problems; an instance may belong to more than one class, and these classes are organized into a hierarchical structure. Real world data are often complex like this. Hierarchical multi-label text classification is becoming ever more popular nowadays, because hierarchical structure can be applied to describe the relationship of textual data. Textual data which we have seen every day are web pages. As the size of web pages has been becoming extremely large, website such as Web directory and Wikipedia need the automated system to classify new web pages in their databases. This kind of problem is, therefore, a large-scale hierarchical multi-label classification. Many researches proposed various methods to deal with the problem, but these methods cannot process large-scale data. The methods may require a large storage space, may take too long to process or may have low accuracy. Meanwhile, some methods that can process large-scale data do not utilize the hierarchical structure at all. This thesis proposed large-scale hierarchical multi-label text classification method that improved k-nearest neighbor method and utilized the hierarchical structure by trained SVM at the top level of hierarchy in order to increase the precision. Furthermore, we removed features that rarely appeared in training dataset to reduce large number of features, and used important features of test data to select training data in order to reduce large number of data. The evaluation showed that our proposed method ranked fourth on Wiki-Medium dataset with 25.70% LBMaF and ranked second on Wiki-Large dataset with 23.48% LBMaF.