Pretrained Language Models’ Effectiveness in Arabic Medical Inquiries Classification
IEEE Access · 2026 · 14 · 96602-96615
DOI: 10.1109/access.2026.3707308 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Mohammed Bahbib und 1 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Online medical services are growing rapidly, creating a significant need for accurate NLP models that can efficiently handle patient needs. Medical question classification is particularly valuable, as it can help route patients to relevant information while supporting medical professionals in managing high query volumes. Existing classification models have shown promising results, yet they still struggle with the morphological complexity of Arabic, domain-specific medical terminology, and the wide diversity of dialectal writing styles across Arabic-speaking regions. Additionally, Arabic Pre-trained Language Models’ (PLMs) effectiveness is still unexplored in the classification of medical questions. To this end, this paper investigates the effectiveness of several Arabic PLMs in classifying Arabic medical questions, using two Arabic medical datasets containing 34,000 and 40,000 questions for multiclass and binary classification, respectively. Each PLM is evaluated using four metrics: accuracy, precision, recall, and F1 score across three random seeds, while the best-performing model is further examined through error analysis and threshold calibration across both datasets. Across the multiclass dataset, scores range from 0.6945 to 0.7474 across all PLMs, while all PLMs perform noticeably better on the binary classification task, with scores ranging from 0.8294 to 0.8757, likely reflecting the simpler decision boundary of a two-class problem. Particularly, AraBERT consistently achieves the best results across both datasets, with SaudiBERT and MarBERT performing competitively on the binary task, consistent with the dominance of MSA and Gulf Arabic writing styles. The obtained results demonstrate the need for Arabic medical PLMs trained on diverse dialectal datasets to better reflect real-world patient language, ultimately improving their reliability in clinical NLP applications.
Zitieren
APA
Bahbib, M., & Yakhlef, M. B. (2026). Pretrained Language Models’ Effectiveness in Arabic Medical Inquiries Classification. IEEE Access, 14, 96602-96615. https://doi.org/10.1109/access.2026.3707308
Vancouver
Bahbib M, Yakhlef MB. Pretrained Language Models’ Effectiveness in Arabic Medical Inquiries Classification. IEEE Access. 2026;14:96602-96615. doi:10.1109/access.2026.3707308
BibTeX
@article{Bahbib2026,
author = {Bahbib, Mohammed and Yakhlef, Majid Ben},
title = {Pretrained Language Models’ Effectiveness in Arabic Medical Inquiries Classification},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {96602--96615},
doi = {10.1109/access.2026.3707308}
}
EUVIMED