Flat and Nested Named Entity Recognition in Arabic Language
Taoufiq El Moussaoui, Chakir Loqman, Jaouad Boumhidi · 2024
Arabic Named Entity Recognition (NER) is re-garded as a difficult problem due to the limited resources, the rich morphology of the language, and its ambiguity. Arabic NER research is frequently limited to flat entities, ignoring the nested ones. Nested entities consist of entities that contain references to other named entities. The research reported in this paper provides an advanced Arabic Named Entity Recognition (ANER) approach for identifying and recognizing both types of entities (flat and nested). We employed a biaffine model to provide our model with a global perspective of the input by utilizing the concept of graph-based dependency parsing. We conducted ex-periments on the ANERCorp dataset, the combined ANERCorp and AQMAR datasets, and the Wojood dataset. We achieved high recognition performance on the three datasets, with an f1 score of 87.30%, 88.98%, and 89.21 % on ANERCorp, merged ANERCorp-AQMAR and Wojood, respectively. The results of this study offer significant insights for future research and further the development of Arabic NER approaches.