Authorship Attribution of Short Historical Arabic Texts Based on Lexical Features

A Comparative Study of Machine Learning Methods for Genre Identification of Classical Arabic Text

“…For Arabic genre identification, there are no previous studies to compare to, however, comparing to Arabic text classification literature our results are also reasonable, as the accuracy of Arabic text classification ranges between 61% and 98% [Alabbas, Al-Khateeb, Mansour et al (2016)]. Comparing our results to attribution studies, our results are comparable to those reported in the literature, Howedi et al [Howedi and Mohd (2014)] report 76.67% accuracy for SVM, and [Ouamour and Sayoud (2012); Ouamour and Sayoud (2013)] indicate performance of 80% by an SVM variant. The least recorded performance was exhibited by the NB classifier at 49% using 70, and 80 MFWs.…”

“…Eder's delta distance has shown the best general attributive success among the four tested distance measures at 75%. Comparing to results reported in the literature, our findings are excellent, as the maximum for distance based classifiers is 60% as reported in Ouamour et al [Ouamour and Sayoud (2013)]. From the comparison of genre identification accuracy for all eight models, the Support Vector Machines (SVM) yielded the highest overall percentage of correct genre attribution (80% correct attributions) among varying MFWs setting from 50 to 500.…”

“…The results indicate a classification accuracy of 80%. Using the same dataset (short historical texts), another study [Ouamour and Sayoud, (2013)] present a comparative study of seven supervised methods for attribution: distance based classifiers including Manhattan distance, Cosine distance, Stamatatos distance, and Canberra distance, a Multi Layer Perceptron (MLP) classifier, the SMO-SVM classifier and a linear regression classifier. The best results are obtained by SMO-SVM at 80% accuracy.…”

An extensive study of the Bag-of-Words approach for gender identification of Arabic articles

“…To the best of our knowledge the authorship attribution problem for the Arabic language has not been studied well as the number of published works is very small [1], [25], [26], [27]. One of the most notable works is that of Abbasi and Chen [1] in which the authors used different sets of features including lexical, syntactic, structural and content-specific features for the authorship attribution problem under both the English and the Arabic languages.…”

“…They tested their approach on a dataset of 14 Arabic novels by six different writers. Ouamour and Sayoud [27] considered a dataset of 30 historic texts written by 10 different authors. What is special about this dataset is that the authors are all famous well-educated Arab explorers describing their travels and expeditions to different regions of the world.…”