Leveraging Hybrid Adaptive Sine Cosine Algorithm with Deep Learning for Arabic Poem Meter Detection

Abstract

Poetry is a significant aspect of any language. Many cultures and the history of nations are recognized in poems. Compared to prose, each poem has a rhythmic structure that is quite different. The language has its set of lyrical structures for poems, known as meters. Detecting the meters of Arabic poems is a complicated and lengthy procedure. The text must be encrypted using the Arudi method to classify the poem's meter, which requires complex rule-based transformation before another set of rules classifies the meters. Applying deep learning (DL) to meter classification in Arabic poems includes constructing a neural network to discern rhythmic patterns inherent in various meters. The model can extract essential features, like word lengths or syllable patterns, by tokenizing and preprocessing text datasets. Architectures such as Long Short-Term Memory Networks (LSTM) or Recurrent Neural Networks (RNNs) are fitting solutions to capture temporal relations in poetic verses. This research introduces a Hybrid Meta-heuristics with Deep Learning for the Arabic Poem Meter Detection and Classification (HMDL-APMDC) model. The main intention of the HMDL-APMDC system is to recognize various kinds of meters in Arabic poems. The HMDL-APMDC technique primarily preprocesses the input dataset to make it compatible with the classification process. Besides, the HMDL-APMDC technique applies Convolution and Attention with a Bi-directional Gated Recurrent Unit (CAT-BiGRU) for the automated recognition of meter classes. Furthermore, the adaptive sine-s-cosine particle swarm optimization (ASCA-PSO) algorithm is applied to optimize the hyperparameter tuning of the CAT-BiGRU model, enhancing the meter detection results. A detailed simulation analysis is made to highlight the improved performance of the HMDL-APMDC technique. The empirical outcomes stated that the HMDL-APMDC technique had a superior outcome of 98.53% over recent models under the MetRec dataset.