Kernels and Similarity Measures for Text Classification

Measuring similarity between two strings is a fundamental step in text classification and other problems of information retrieval. Recently, kernel-based methods have been proposed for this task; since kernels are inner products in a feature space, they naturally induce similarity measures. Information theoretic (dis)similarities have also been the subject of recent research. This paper describes some string kernels and information theoretic mesures and shows how they can be efficiently implemented via suffix trees. The performance of these measures is then evaluated on a text classification (authorship attribution) problem, involving a set of books by Portuguese writers.