Automatic Identification of Poet in Arabic Poetry

Abstract

Identification of a poet, for a given poem, is the task of distinguishing authority of poetic texts among different authors. It is a difficult process because it requires knowledge in Arabic language and poetry with semantic relation and poet fingerprint. In this thesis, a system was created that extract prosody style, the poetic meter and recognizes the poet (author) of a given Arabic poem automatically. For doing this, a collection of 2044 labeled poems to seven poets was collected using private crawler and private cleaning program. Firstly, the poem is transformed into Prosody styles using Prosody rules. Then the poem meter will be extracted using Naïve Bayes and Decision Tree classifiers that learned from the data set. The process of identifying the poet is carried out in three methodologies: using content, features and combination of them with poem meter. The process of identification of poet of a given poem is done using four classification techniques; Naïve Bayes, Support Vector Machine, k Nearest Neighbor, and Decision Tree.  The obtained results for identifying the poem meter were 0.723 and 0.735 using NB and DT classifier respectively. The best obtained result for identifying poets of poems, using combination of content and feature with support of poem meter, was 0.797, using NB, classifier. The results showed the identification of poet for a given poem is attainable with acceptable precision. Moreover, some algorithms are not suitable for this task.