A Genetic Approach to Chinese Class Descriptor Generation

Abstract

To facilitate management of a huge amount of electronic data, efficient classification is needed. One critical issue to affect classification is the selection of class descriptors. In this paper a genetic-based class descriptor generator is proposed and investigated for Chinese document classification. It is based on the idea that with a well-designed genetic algorithm, a set of fitter descriptors can be obtained. Meanwhile, a content-based finder is incorporated with the proposed generator to discover the initial set of chromosomes. The proposed generator is verified with a real corpus in both unweighted and weighted settings. Comparisons to the traditional union-based approach are also made and the experimental results show that the generator indeed produces a better set of class descriptors than union-based model in both the cases of predefined classes and re-clustered classes.