Features selection for text classification based on constraints for term weights

Text classification is an important data analysis problem which can be applied in different domains including airspace industry. In this paper different text classification problems such as opinion mining and topic categorization are considered. Different text preprocessing techniques (TF-IDF, ConfWeight, and the Novel TW) and machine learning algorithms for classification (Bayes classifier, k-NN, SVM, and artificial neural network) are applied. The main goal of the presented investigations is to decrease text classification problem dimensionality by using features selection based on constraints for term weights. Such features selection provides significant reduction of dimensionality and less computational time for calculations. Besides, the use of constraints for term weights could increase classification effectiveness. We have observed such increase for three out of five problems. In the remaining two problems, no significant change and a decrease of classification effectiveness was observed.