**7. Conclusions**

Arabic is a challenging language because of its high morphological ambiguity, writing style, and absence of capitalization. Thus, any NLP task for this language requires a large number of feature engineering and preprocessing steps. In this study, we experiment with a B-RNN with LSTM/GRU for the ANER task. Without using any feature engineering or additional preprocessing, we tackle the problem of NER for the Arabic text. We find that deep learning-based approaches, especially LSTM network, are useful for identifying Arabic NEs and can efficiently outperform many other methods based on manually engineered features or rule systems. The incorporation of pre-trained word embedding enables the system to obtain considerable improvements in the recognition task and achieve excellent results in F-score measures, where we achieved a high F-score measure of approximately 88.01% and 87.12% for Bi-LSTM and Bi-GRU respectively.

**Author Contributions:** M.N.A.A. wrote the paper with support from A.H. A.H carried out the experiments with the help of M.N.A.A. G.T. supervised the whole project. All the authors revised the manuscript.

**Funding:** This work was partially funded by the National Natural Science Foundation of China (Project No. 61403422).

**Conflicts of Interest:** The authors declare no conflicts of interest.
