DarNERcorp: An annotated named entity recognition dataset in the Moroccan dialect

Abstract

DarNERcorp is a manually annotated named entity recognition (NER) dataset in the Moroccan dialect, also called Darija. The dataset consists of 65,905 tokens and their corresponding tags according to BIO scheme. 13.8% of the tokens are named entities spanning four categories: person, location, organization, and miscellaneous. The data were scraped from the Moroccan Dialect section of Wikipedia and processed and annotated using open-source libraries and tools. The data are useful for the Arabic natural language processing (NLP) community as they address the lack in dialectal Arabic annotated corpora. This dataset can be used to train and evaluate named entity recognition systems in dialectal and mixed Arabic.

Keywords

• Natural language processing
• Text mining
• Named entity recognition
• Dialectal Arabic
• Corpus
• BIO