Authorship attribution in Arabic poetry

Abstract

In this paper, we present the Arabic poetry as an authorship attribution task. Several features such as Characters, Sentence length; Word length, Rhyme, and First word in sentence are used as input data for Markov Chain methods. The data is filtered by removing the punctuation and alphanumeric marks that were present in the original text. The data set of experiment was divided into two groups: training dataset with known authors and test dataset with unknown authors. In the experiment, a set of thirty-three poets from different eras have been used. The Experiment shows interesting results with classification precision of 96.96%.