A Comparative Study for Yemeni Poets Detection Using TEXT-CNN and RNN-LSTM Text Classification

Abstract

The growing prevalence of internet usage has led to a substantial capacity in textual data. Text classification is an essential field in natural language processing (NLP). It differs in various domains and languages. This study aims to classify Arabic poetry to identify the poets, with a particular focus on Yemeni poets, who wrote in Modern Standard Arabic (MSA). We propose an approach that merges standard text pre-processing techniques with a hybrid algorithm, by combining the Levenshtein and Jaccard methods, to improve poet identification. To the best of our knowledge, this research is the first to utilize deep learning techniques for identifying Yemeni poets. It preprocesses Arabic text to enhance name matching, to enable the compilation of poets' verses despite variations in name presentation. Besides, two deep learning models were constructed and their performance was assessed. The deep learning architectures employed include TEXT-CNN and RNN-LSTM; these were evaluated using a range of learning rate values alongside other fixed hyperparameters. The models were trained on poetry texts utilizing two main text representation methods: word embedding through Word2Vec and document embedding via Doc2Vec, which significantly improved the training and testing efficiency, for poet recognition. The findings indicated that the TEXT-CNN model exceeded the RNN-LSTM model in training and testing accuracy. However, the RNN-LSTM results exhibited greater robustness and generalizability.