Akram Roshdi - Department of Computer, Islamic Azad University, Khoy Branch,Iran

Today, with the increasingly growing volume of textinformation, text classification methods seem to be essential.Also, increase in the volume of Persian text resources adds tothe importance of this issue. However, classification workswhich have been especially done in Persian are not still asextensive as those of Latin, Chinese, etc. In this paper, a systemfor Persian text classification is presented. This system is able toimprove the standards of accuracy, retrieval and total efficiency.To achieve this goal, in this system, after texts preprocessingand feature extraction, a new improved method of featureselection based on Particle Swarm Optimization algorithm(PSO) is innovated for reducing dimension of feature vector.Eventually, the classification methods are applied in the reducedfeature vector. To evaluate feature selection methods in theproposed classification system, classifiers of support vectormachine (SVM), Naive Bayes, K nearest neighbor (KNN) andDecision Tree are employed. Results of the tests obtained fromthe implementation of the proposed system on a set ofHamshahri texts indicated its improved precision, recall, andoverall efficiency. Also, SVM classification method had betterperformance in this paper.

Feature vector, classification,support vectormachines, Feature Extraction, Dimensions Reduction