Classical Arabic Named Entity Recognition Using Variant Deep Neural Network Architectures and BERT

Recurrent Neural Networks (RNNs) and transformers are deep learning models that have achieved remarkable success in several Natural Language Processing (NLP) tasks since they do not rely on handcrafted features nor enormous knowledge resources. Named Entity Recognition (NER) is an essential NLP task used many applications such as information retrieval, question answering, machine translation. NER aims to locate, extract, classify named entities into predefined categories person, organization location. Arabic considered a challenging because of the complexity unique characteristics Arabic. Most previous research based-Arabic focused Modern Standard Dialectal Arabic, which different variations from Classical In this paper, we investigate learning-based using neural network architectures BERT based contextual language model trained general domain text. We propose two RNN-based by fine-tunning pretrained recognize The pre-trained representations were input BGRU/BLSTM fine-tuned dataset. addition, explore variant proposed BERT-BGRU/BLSTM-CRF models. Experimentations showed BERT-BGRU-CRF outperformed other achieving F-measure 94.76% CANERCorpus. To best our knowledge, first work learning.