Study On Techniques Of Rough Set Based SVM Hierachical Categorization

The data processed in text categorization are complicated and frequently updated. Accordingly, heirachical classification and incremental learning are in-depth studied in this thesis.Learned by the analysis of attribute reduction algorithm idea in the rough-set theory, a method based on the rough-set theory is brought forward. The method can refrain the text computations from curses of dimensionality. After the feature selection, a method for hierarchical categorization is given, based on the virtual tree structure. Firstly, this method adjusts the unbalanced samples with the pre-selecting strategy, in order to avoid over fitting in larger classes. Secondly, the upper-layer classifiers are designed to be multi-labled, so as to reduce the blocking which negatively affected the lower ones. Finally,a threshold control strategy is used to reduce the information losing while sifting the history samples in the incremental learning.In the last chapter of the thesis, the design and implementation of a heirachical categorization system is introduced. The experiment results show its feasibility and validity over the traditional methods.