Trained SVMs based rules extraction method for text classification

摘要

The automatic text classification method aims to assign text files to one or more predefined categories according to the text information contained by all kinds of text format files.SVM is recognized as one of the most effective text classification methods for its high accuracy, but its black-box feature causes that the description of each category can not be given and explained. In this paper, a new rule extraction method for text classification based on trained SVMs is proposed to solve the bottleneck of SVMs. The Experiments show that the proposed approach can improve the validity of the extracted rules remarkably compared to C4.5 either in speed or accuracy.