Using One-Class SVMs for Relevant Sentence Extraction

Canasai Kruengkrai and Chuleerat Jaruskulchai

Abstract

In this paper, we present a novel approach to relevant sentence extraction, especially using only positive examples for training. Our approach applies a methodology of Support Vector Machines (SVMs) for one-class classification called one-class SVMs. The idea is to transform the data into the feature space corresponding to the kernel, and then separate them from the origin with maximum margin. We also examine a method for analyzing on a subset of features that appears to be the best discriminate indicators. Experiments on two different text corpora, including newspaper articles and technical papers, show that our approach gives reasonable results.

Download: pdf, ps

Canasai Kruengkrai