Elamur Research⌕
☰
home / papers / 2026 / HCMUS_PrompterXPrompter at AbjadMed: When Classification Meets Retrieval: Taming the Long Tail in Arabic Medical Text Classification

HCMUS_PrompterXPrompter at AbjadMed: When Classification Meets Retrieval: Taming the Long Tail in Arabic Medical Text Classification

Paper · 2026 · Duy Minh Dao Sy, Trung Kiet Huynh, Nguyen Dinh Ha Duong et al.

Summary

This paper describes a system for the ABJAD-NLP Shared Task at EACL 2026. The task is to sort Arabic medical consultations, written as question–answer pairs, into 82 topic labels. The data is very unbalanced: the largest class has 600 examples and the smallest only 7. Because scoring uses macro-F1, rare labels count as much as frequent ones.

The authors' system, HybridMed, works in three steps:

  • Preprocessing: every training item contains explicit question and answer markers, so the answer marker is replaced with a [SEP] token.
  • Classification: MARBERTv2 is fine-tuned as a classifier.
  • Retrieval: the same model's [CLS] vectors index the training set in FAISS. For each test item, the labels of its 10 closest training examples are blended with the classifier's probabilities.

HybridMed scored 0.4902 macro-F1 on the private test set, ahead of more than ten alternatives the team tried. The main lesson is that simple nearest-neighbour lookup can help a neural model on classes with very few examples.

Key points

  • Data: 27,951 training and 18,634 test items. 24 of the 82 classes have fewer than 50 examples, and the gap between the largest and smallest class is roughly 85:1.
  • Model choice: in a pilot comparison, MARBERTv2 (0.3531) beat CAMeLBERT-Mix (0.3445) and AraBERT (0.3312).
  • Blending weight: the classifier gets weight α = 0.7 and the k-NN retrieval (k=10) gets 0.3. This value was picked by grid search on a stratified 90/10 validation split.
  • Ablation on validation F1: the question–answer preprocessing added about 5 points and retrieval about 2 more. The authors estimate retrieval improved the tail classes by about 8%.
  • Other systems tested: Monte Carlo Dropout reached 0.4633 private F1 and BERT + XGBoost reached 0.4611. Contrastive learning and a simple ensemble stayed near 0.22.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksInformation retrieval & search · 473 works

Cite

Cite as Duy Minh Dao Sy, Trung Kiet Huynh, Nguyen Dinh Ha Duong, Nguyen Chi Tran, Phu Quy Nguyen Lam and Hoa Pham Phu. 2026. HCMUS_PrompterXPrompter at AbjadMed: When Classification Meets Retrieval: Taming the Long Tail in Arabic Medical Text Classification. Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.abjadnlp-1.7

@inproceedings{sy2026hcmus,
  title = {{HCMUS\_PrompterXPrompter at AbjadMed: When Classification Meets Retrieval: Taming the Long Tail in Arabic Medical Text Classification}},
  author = {Duy Minh Dao Sy, Trung Kiet Huynh, Nguyen Dinh Ha Duong, Nguyen Chi Tran, Phu Quy Nguyen Lam and Hoa Pham Phu},
  booktitle = {Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script},
  year = {2026},
  doi = {10.18653/v1/2026.abjadnlp-1.7},
  url = {https://doi.org/10.18653/v1/2026.abjadnlp-1.7}
}

References

11 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Kuniaki Sohn, David Berthelot, Chun-Liang Li, Zizhao Zhang, Nicholas Carlini, Colin Raffel, Tom Sercu, and Ruchir Goswami. 2020. Decoupled pseudo-labeling for semi-supervised learning. Advances in Neural Information Processing Systems, 33:21798– 21810. find
  2. Muhammad Abdul-Mageed, AbdelRahim Elmadany, and El Moatez Billah Nagoudi. 2021. ARBERT & MARBERT: Deep bidirectional transformers for Arabic. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 7088–7105, Online. Association for Computational Linguistics. ↗
  3. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, et al. 2020. Hugging face’s transformers: State-of-the-art natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 38–45.
  4. Wissam Antoun, Fady Baly, and Hazem Hajj. 2020. Arabert: Transformer-based model for arabic language understanding. In Proceedings of the 4th Workshop on Open-Source Arabic Corpora and Processing Tools, pages 9–15. ↗
  5. Tianqi Chen and Carlos Guestrin. 2016. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 785–794. ↗
  6. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4171–4186.
  7. Yarin Gal and Zoubin Ghahramani. 2016. Dropout as a Bayesian approximation: Representing model uncertainty in deep learning. In Proceedings of the 33rd International Conference on Machine Learning, pages 1050–1059. ↗
  8. Pranav Gupta, Niranjan Kumar M, Balaji Nagarajan, Imed Zitouni, and Mo El-Haj. 2026. Abjadmed: Arabic medical text classification at abjadnlp 2026. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco.
  9. Go Inoue, Bashar Alhafni, Nurpeiis Baimukan, Houda Bouamor, and Nizar Habash. 2021. The interplay of variant, size, and task type in Arabic pre-trained language models. In Proceedings of the Sixth Arabic Natural Language Processing Workshop, pages 92– 104, Kyiv, Ukraine (Virtual). Association for Computational Linguistics.
  10. Jeff Johnson, Matthijs Douze, and Hervé Jégou. 2019. Billion-scale similarity search with GPUs. IEEE Transactions on Big Data, 7(3):535–547. ↗
  11. Urvashi Khandelwal, Omer Levy, Dan Jurafsky, Luke Zettlemoyer, and Mike Lewis. 2020. Generalization through memorization: Nearest neighbor language models. In International Conference on Learning Representations. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sy-2026-hcmus-prompterxprompter-abjadmed-when-classification/