Improved single-label text categorization by instance filtration

Abstract

Machine learning classifiers are widely used for text categorization however a classifier misclassifies some of the instances into a category that is relevant to their actual category. The categorization ability of a classifier can be improved by filtering dataset with better classifier and removing such category for misclassified instances. In this paper we proposed a two level approach where level-1 filters instances according to their likelihood in each category and reduce training dataset to top ranked 't' categories and their instances whereas level-2 classifier is used to classify instances with filtered training set. We employed Naïve Bayes, SVM and KNN as machine learning classifiers. Experimental evaluations on standard reuters-21578 and 20 Newsgroups datasets showed improved categorization effectiveness as measured by accuracy, precision, recall and f-measure protocols.