Multiple Support Vector Machines for Binary Text Classification Based on Sliding Window Technique

Abstract

Supervised machine learning algorithms, such as support vector machines (SVMs), are widely used for solving classification tasks. In binary text classification, linear SVM has shown remarkable efficiency for classifying documents due to its superior performance. It tries to create the best decision boundary that enables the separation of positive and negative documents with the largest margin hyperplane. However, in most cases there are regions in which positive and negative documents are mixed due to the uncertain boundary. With an uncertain boundary, the learning classifier is more complex, and it often becomes difficult for a single classifier to accurately classify all unknown testing samples into classes. Therefore, more innovative methods and techniques are needed to solve the uncertain boundary problem that was traditionally solved by non-linear SVM. In this paper, multiple support vector machines are proposed that can effectively deal with the uncertain boundary and improve predictive accuracy in linear SVM for data having uncertainties. This is achieved by dividing the training documents into three distinct regions (positive, boundary, and negative regions) based on a sliding window technique to ensure the certainty of extracted knowledge to describe relevant information. The model then derives new training samples to build a multiple SVMs based classifier. The experimental results on the TREC topics and standard dataset Reuters Corpus Volume 1 (RCV1), indicated that the proposed model significantly outperforms six state-of-the-art baseline models in binary text classification.