Deep Learning Techniques for Identifying Poets in Arabic Poetry: A Focus on LSTM and Bi-LSTM
Hamza Shoubaki, Sherief Abdallah, Khaled F. Shaalan · Procedia Computer Science · 2024
In this paper, we introduce a comprehensive approach to the classification of Arabic poetry using deep learning techniques. We utilized a dataset comprising nearly one million records of Arabic poetry verses, each labeled with nine different poets encompassing both classical and modern poetic styles. We explored various algorithms to identify the most effective models for accurately determining the poet's identity from the verses. Based on our analysis, we selected LSTM (Long Short-Term Memory) and Bi-LSTM (Bidirectional LSTM) as our primary baseline models, given the demonstrated efficacy of RNN (Recurrent Neural Network) variants in text classification. LSTM, in particular, has shown significant proficiency in analyzing sequential data across multiple languages. Our results indicate a promising average classification accuracy of 92.35%, highlighting the potential for automating the classification of texts in morphologically complex languages such as Arabic. Notably, Bi-LSTM marginally outperformed the standard LSTM, achieving average accuracy of 92.56%. We further discuss the implications of our findings for Arabic literature, particularly in the realm of poetry, and address the challenges encountered during the study. These findings represent a significant advancement in Arabic NLP, offering a powerful framework for poet identification and paving the way for future applications of deep learning in processing Arabic literary texts.