BERT-AraPeotry: BERT-based Arabic Poems Classification Model
Abdulfattah E. Ba Alawi, Ferhat Bozkurt, Mete Yağanoğlu · 2024
Based on the time of writing, Arabic poems are divided into eras or periods. Each era has its poetic properties that characterize its poems. The classification of poems and literary texts into eras using conventional Natural Language Processing (NLP) techniques and computerized methods is a challenging task. To overcome this problem, this paper uses a modified version of the Bidirectional Encoder Representations from Transformers (BERT) model to classify Arabic poems. The proposed model was modified to better comprehend and classify the rich grammatical and stylistic features of Arabic poetry. We modified the Multilingual Arabic variants of BERT (MARBERT) by adding pooling, dropout, and classification layers and training it on a large corpus of Arabic poetry from different genres. Additionally, linguistic and rhyme features were employed to enhance the performance. The objective of this method is to offer the model a sophisticated comprehension of the theme and poetry phrases in Arabic poems. The suggested model performs more effectively in categorizing Arabic poetry with an accuracy of 86.82%. The findings illustrated that Arabic literary text analysis and classification can be automated and emphasized the significance of language-specific optimization for transformer-based models designed to tackle challenging categorization of textual data.