Transformer-Based Method for Arabic Entity Extraction

The Internet is witnessing a significant increase in textual information, especially in the Arabic language, which makes extracting and classifying entities such as individuals, institutions, or sites within the text necessary for more than one application, such as answering questions or analyzing feelings, and other various applications. Processing Arabic texts and identifying entities named in Arabic (ANER) in Arabic text represents a great challenge because the Arabic language is characterized by complexity that differs from other languages, such as the grammatical and morphological structure of the language, the diversity of its dialects, and its vocabulary that carries multiple meanings depending on the context, which increases the complexity. Entity recognition. Therefore, this paper proposes a new model for Arabic Named Entity Recognition (NER) that overcomes the challenges faced by Arabic word processors. This model relies on transformers, specifically XLM-R, for Arabic Named Entity Recognition (NER). Model performance was rigorously evaluated on the XTREME dataset (WikiANNIPAN-X), where this methodology includes advanced preprocessing using SentencePiece and fine-tuning of the Transformer model. Experimental results indicate a validation loss of 0.1920, an F1 score of 0.8882, and an accuracy of 0.9538, which confirms the accuracy and effectiveness of the model in NER tasks. This work takes advantage of the comprehensive multilingual capabilities of the XLM- R converter. It contributes to the development of Arabic NER processes, and measures the significant improvement in dealing with low-resource languages within NLP applications for detailed foundational methodologies.

Hali tarjima qilinmagan