Arabic Text Categorization

In this paper, we compare the performance of three classifiers for Arabic text categorization. In particular, the naïve Bayes, k-nearest-neighbors (knn), and distance-based classifiers were used. Unclassified documents were preprocessed by removing punctuation marks and stopwords. Each document is then represented as a vector of words (or of words and their frequencies as in the case of the naïve Bayes classifier). Stemming was used to reduce the dimensionality of feature vectors of documents. The accuracy of the classifiers is compared using recall, precision, error rate and fallout. The results of the experimentations that were carried out on an in-house collected Arabic text show that the naïve Bayes classifier outperforms the other two.