Combining LSI with other Classifiers to Improve Accuracy of Single-label Text Categorization

Abstract—This paper describes the combination of k-NN and SVM with LSI to improve their performance in single-label text categorization tasks, and the experiments performed with six datasets to show that both k-NN-LSI (the combination of k-NN with LSI) and SVM-LSI (the combination of SVM with LSI) outperform the original methods for a significant fraction of the datasets. Overall, both combinations present an average Accuracy over the six datasets used in this work that is higher than the average Accuracy of each original method. Having in mind that SVM is usually considered the best performing classification method, it is particularly interesting that the combinations perform even better for some datasets. I. INTRODUCTION AND EXPERIMENTAL SETTING The main goal of text categorization (TC) is to derive methods for the categorization of natural language text. Th