An Efficient SVM Classifier for Lopsided Corpora

摘要

This paper explores application of SVM to lopsided-corpora in text categorization. By means of integrating kernel caching with shrinking policies effectively, an improved SVM training algorithm for weight-calculation formula is proposed under the decomposition framework. Extensive experiments on lopsided-corpora have been conducted. The conclusion can make it possible to apply the improved SVM training algorithm to lopsided corpora in text categorization