Systematic Literature Review of Arabic NLP Datasets: A Meta-Study

This meta-analysis is a systematic review and synthesis of the Arabic Natural Language  Processing (NLP) dataset landscape, in accordance with PRISMA guidelines. The  review locates and categorizes publicly accessible datasets in NLP tasks that span a  spectrum of sentiment analysis, text classification, question answering, summarization,  dialect detection, and paraphrasing. Key datasets catalogues like Masader and Masader  Plus are underscored as driven by enhancing discoverability and metadata  standardisation, whereas task and domain-specific creation of resources are represented  by ArabSis, SNAD, A-MASA, AGS and WiHArD. As can be seen, although the  amount and variety of the datasets have grown in recent years, there is still a  considerable number of gaps in the coverage of dialects, domain specificity and quality  of annotations. The review ends with suggestions on how to extend underexploited  areas, annotation behavior, and open-access to accelerate the Arabic NLP research and  application.