Scholarlink

<p>This survey reviews advancements in Arabic Word Sense Disambiguation (WSD), with a focus on contextualized methods such as deep learning and transformer models. Despite the linguistic complexities of Arabic and the scarcity of annotated datasets, WSD remains a critical challenge in Natural Language Processing (NLP). This paper systematically examines traditional knowledge-based methods, machine learning approaches, deep learning models, and hybrid techniques, with a particular emphasis on transformer models such as BERT, which have recently achieved state-of-the-art performance in Arabic WSD. The survey outlines the criteria for model selection, including performance on benchmark datasets and relevance to real-world applications. It also provides a comparative evaluation of the methodologies, highlighting their strengths and weaknesses through structured comparisons of Accuracy, Precision, Recall, and F1-score. Unlike previous surveys, this work offers a comprehensive analysis that integrates multiple evaluation metrics across diverse studies. The paper concludes by identifying key challenges, such as the need for larger annotated datasets and improved contextual understanding, while proposing future research directions, including domain adaptation and cross-dialectal WSD.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Arabic word sense disambiguation: a survey in the era of transformer-based models

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Rakia Saidi,
  • Fethi Jarray

摘要

This survey reviews advancements in Arabic Word Sense Disambiguation (WSD), with a focus on contextualized methods such as deep learning and transformer models. Despite the linguistic complexities of Arabic and the scarcity of annotated datasets, WSD remains a critical challenge in Natural Language Processing (NLP). This paper systematically examines traditional knowledge-based methods, machine learning approaches, deep learning models, and hybrid techniques, with a particular emphasis on transformer models such as BERT, which have recently achieved state-of-the-art performance in Arabic WSD. The survey outlines the criteria for model selection, including performance on benchmark datasets and relevance to real-world applications. It also provides a comparative evaluation of the methodologies, highlighting their strengths and weaknesses through structured comparisons of Accuracy, Precision, Recall, and F1-score. Unlike previous surveys, this work offers a comprehensive analysis that integrates multiple evaluation metrics across diverse studies. The paper concludes by identifying key challenges, such as the need for larger annotated datasets and improved contextual understanding, while proposing future research directions, including domain adaptation and cross-dialectal WSD.