Designing A Hybrid Deep Learning Methodology for Arabic Text Classification

The process of automatically assigning relevant labels to a corpus of text is known as text categorization. Single-label text categorization is used when there is only one label to use when sorting the content. There are challenges with the multi-label form, however. The lack of large publicly accessible Arabic rich and rational datasets complicates both of these endeavours. Therefore, we provide novel, comprehensive, and objective Arabic text categorization datasets, both multi-label (NADiA) & single-label (SANAD). Both corpora are freely available for use by researchers in the area of computational linguistics applied to the Arabic language. We provide a Hybrid Deep Learning (HDL) method for text in Arabic classification using Long-Short Term Memory (LSTM) & Gated Recurrent Neural Network (GRU), which may be used to assess the efficacy of such models on SANAD and NADiA. In contrast to previous studies, our method relies only on deep learning algorithms and does not include any pre-processing. We also attempted to use word2vec embed models to enhance classification precision. The largest number of groups that attention-GRU correctly classified on NADiA's "Masrawy" database was ten, and its average overall accuracy was 88.68%.