Advanced Deep Learning Approach for Meter Classification in Arabic Poetry Using Bidirectional LSTM Networks

Abstract

This dissertation develops and evaluates an advanced artificial intelligence system designed to classify Arabic poems based on their metrical patterns, using a Bidirectional Long Short- Term Memory (Bi-LSTM) network. Given the linguistic complexity of Arabic, which includes extensive morphological variations and rich phonetic patterns, this study addresses significant challenges in the automated classification of poetic meters. The research employs a robust methodology involving advanced text preprocessing techniques such as tokenization, sequence padding, and label encoding to prepare a comprehensive dataset for machine learning. The model is trained and optimized on Google Colab's TPU resources, which enhances computational efficiency and expedites the iterative refinement process. The effectiveness of the system is meticulously assessed through a variety of metrics, including accuracy, precision, recall, and F1-score, to ensure a thorough understanding of its performance across different poetic meters. Validation and testing phases are incorporated to evaluate the model's generalization abilities, utilizing confusion matrices and early stopping mechanisms to pinpoint areas for potential improvement. The findings demonstrate that Bi-LSTM networks are particularly effective in handling the complexities associated with Arabic poetic texts. This project advances the practical application of automatic poetic meter classification and encourage further scholarly exploration of deep learning techniques within Arabic literary studies. Ultimately, this dissertation highlights the potential of AI to enhance the accessibility and analytical depth of Arabic poetry, enriching the appreciation and understanding of this venerable literary tradition.