Improving the Performance of Naïve Bayes Algorithm for Arabic Text Categorization

• Published in: International Journal of Advanced Studies in Computers, Science and Engineering Vol. 5; no. 11; p. 105
• Main Authors: Al Mashaykhi, Akram M O, Aqoulah, Nibras Jamal Abu, Riadh, May H
• Format: Journal Article
• Language: English
• Published: Gothenburg International Journal of Advanced Studies in Computers, Science and Engineering 01.11.2016
• Subjects: Algorithms; Bayesian analysis; Classification; Classifiers; Mathematical models; Similarity; Texts; Thresholds
• ISSN: 2278-7917

Text Categorization (classification) is the process of classifying documents into a predefined set of categories based on their content, In this paper four techniques are implemented using Naïve Bayes classifier for Arabic text categorization, these techniques are: (TF only ,TF-IDF, Normalized TF-IDF, and N-Gram with N=2 statistical stemmer with threshold similarity 0.8). The four techniques are evaluated by two test set. The results shows that the Normalized TF-IDF and N-Gram with N=2 statistical stemmer with threshold similarity 0.8 technique has the best accuracy ,the Analysis of Naïve Bayes classifier algorithm showed at least two Advantages: first it Work well on numeric and textual data and second its easiness in implementation and computation comparing with other algorithms also the work highlighting at least three Disadvantages: first the Conditional independence assumption is violated by real-world data; second its perform very poorly when features are highly correlated and the last disadvantages it does not consider frequency of word occurrences.