---
title: "DRU at WojoodNER 2024: A Multi-level Method Approach (2024)"
description: "The paper describes the DRU team's entry to the WojoodNER 2024 shared task on fine-grained Arabic named entity recognition. 19 references."
authors: ["Hadi Hamoud", "Chadi Abou Chakra", "Nancy Hamdan", "Osama Rakan Al Mraikhat", "Doha Albared", "Fadi A. Zaraket"]
year: 2024
venue: "Proceedings of the Second Arabic Natural Language Processing Conference"
source: "https://aclanthology.org/2024.arabicnlp-1.104/"
doi: "10.18653/v1/2024.arabicnlp-1.104"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/hamoud-2024-dru-wojoodner-2024-multi-level/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/hamoud-2024-dru-wojoodner-2024-multi-level/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2024](https://md.research.elamur.ai/free/en/papers/2024/page/16/) / DRU at WojoodNER 2024: A Multi-level Method Approach

# DRU at WojoodNER 2024: A Multi-level Method Approach

**Authors:** [Hadi Hamoud](https://md.research.elamur.ai/free/en/authors/hadi-hamoud/), [Chadi Abou Chakra](https://md.research.elamur.ai/free/en/authors/chadi-abou-chakra/), [Nancy Hamdan](https://md.research.elamur.ai/free/en/authors/nancy-hamdan/), [Osama Rakan Al Mraikhat](https://md.research.elamur.ai/free/en/authors/osama-rakan-al-mraikhat/), [Doha Albared](https://md.research.elamur.ai/free/en/authors/doha-albared/) and [Fadi A. Zaraket](https://md.research.elamur.ai/free/en/authors/fadi-zaraket/) · **Year:** 2024 · **Venue:** [Proceedings of the Second Arabic Natural Language Processing Conference](https://md.research.elamur.ai/free/en/venues/arabicnlp/) ([ArabicNLP 2024](https://md.research.elamur.ai/free/en/venues/arabicnlp/2024/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2024.arabicnlp-1.104/) · [PDF](https://aclanthology.org/2024.arabicnlp-1.104.pdf) · [DOI 10.18653/v1/2024.arabicnlp-1.104](https://doi.org/10.18653/v1/2024.arabicnlp-1.104) · [OpenAlex](https://openalex.org/W4402671325)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper describes the DRU team's entry to the WojoodNER 2024 shared task on fine-grained Arabic named entity recognition. Both the flat and the nested sub-task are covered. The data is the Wojood-Fine corpus, which has 21 main entity types. Four of them (GPE, LOC, ORG, FAC) are split further into 31 sub-types, and only the official training and development sets were allowed.

All systems treat NER as multi-label token classification over a one-hot vector of B-/I- tags. They use a sigmoid-based binary cross-entropy loss, so one token can hold several overlapping labels. The models are trained on the nested data, and flat output is produced by keeping the top-level type.

Three designs were compared:
- **Single model:** one model predicts everything, tested with BLOOMZ-7b-mt, GEMMA-7b and AraBERTv2.
- **One×1:** one AraBERTv2 for main types plus one for sub-types.
- **One×4:** one main-type model plus four group-specific sub-type models.

The best was a One×4 variant that moves hybrid sub-types up to the main-type level. It reached F1 0.8782 flat and 0.9043 nested and placed second in the nested track. The work suggests a small encoder with task decomposition can match or beat 7B-parameter LLMs on this kind of labeling.

## Key points
- Training split: 23,125 sentences and 390,999 tokens. Development split: 3,304 sentences and 57,555 tokens. Tags use the IOB scheme, with a 103-dimensional label vector covering types, sub-types and "O".
- Single-model results (flat / nested F1):
  - AraBERTv2: 0.8780 / 0.9040
  - fine-tuned BLOOMZ-7b-mt: 0.8690 / 0.8996
  - GEMMA-7b: 0.8605 / 0.8840
- One×1 slightly underperformed the single model (0.8746 / 0.9033). One×4 experiment 2 was best (0.8782 / 0.9043) and beat the earlier Wojood-Fine nested result of 0.885.
- Per-label scores on dev varied widely:
  - Strong: MED reached F1 ≈ 0.996, and COUNTRY and TOWN were above 0.97.
  - Weak: GPE_ORG reached about 0.59 and WEBSITE about 0.49.
  - Failing: TIME got about 0.05 and CURR got 0.
- Training setup:
  - BLOOMZ and GEMMA: 12 epochs on one A100 80GB, sequence length 512.
  - AraBERTv2: 10 epochs on one A4000 16GB.
  - Special tokens from model tokenizers were masked with the -100 ignore label.

## Interesting ideas
- Replacing a softmax over tags with independent per-label sigmoids plus thresholds lets one token carry several co-existing labels, which is useful wherever categories overlap rather than compete.
- The team trained only on the richest (nested) annotation and derived the simpler flat output with a deterministic rule, so one model serves both views.
- Splitting prediction into a coarse model plus separate fine-grained models per coarse group narrows each model's decision space and reduces confusion between unrelated sub-labels.
- Hybrid categories such as a place acting as an organization were moved up to the main-type level instead of being kept as a tangled sub-type, which removed a source of class-intersection errors.
- A per-label precision/recall table exposed classes with near-zero F1 that a strong overall score hid.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (world): Entity typing against the WordNet/Wikidata ontology could assign several compatible types to one mention at once instead of forcing a single class.
- **I2** (tree): The parser could be trained toward a richer UD layer, such as enhanced dependencies, with the basic tree recovered by an explicit, explainable projection.
- **I3** (frames): Semantic role labelling could first pick a VerbNet class and then call a class-specific role assigner, keeping each decision small and inspectable.
- **I4** (world): Metonymic entities, such as a country standing for its government, could be stored as one entity with two explicit top-level facets rather than as an odd sub-class.
- **I5** (learn): Every evaluation run could report per-label scores and flag collapsed categories as targets for extra silver data or repair rules.

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

## Institutions

- [Arab Center for Research and Policy Studies](https://md.research.elamur.ai/free/en/labs/arab-center-for-research-and-policy-studies/) — Research institute, Doha, Qatar · 9 works · 69 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 3,066 works

## Cite

Cite as: Hadi Hamoud, Chadi Abou Chakra, Nancy Hamdan, Osama Rakan Al Mraikhat, Doha Albared and Fadi A. Zaraket. 2024. DRU at WojoodNER 2024: A Multi-level Method Approach. Proceedings of the Second Arabic Natural Language Processing Conference.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{hamoud2024dru,
  title = {{DRU at WojoodNER 2024: A Multi-level Method Approach}},
  author = {Hadi Hamoud, Chadi Abou Chakra, Nancy Hamdan, Osama Rakan {Al Mraikhat}, Doha Albared and Fadi A. Zaraket},
  booktitle = {Proceedings of the Second Arabic Natural Language Processing Conference},
  year = {2024},
  doi = {10.18653/v1/2024.arabicnlp-1.104},
  url = {https://doi.org/10.18653/v1/2024.arabicnlp-1.104}
}
```

- [← previous DILAB at #SMM4H 2024: RoBERTa Ensemble for Identifying Children's Medical Disorders in English Tweets](https://md.research.elamur.ai/free/en/papers/wasi-2024-dilab-smm4h-2024-roberta-ensemble/)
- [next → Dynamic changes in search strategy during chain-of-thought reasoning of language models](https://md.research.elamur.ai/free/en/papers/aoki-2024-dynamic-changes-search-strategy/)
- [↑ back to papers of 2024](https://md.research.elamur.ai/free/en/papers/2024/page/16/)

## References

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

1. Ahmed Abdelali, Sabit Hassan, Hamdy Mubarak, Kareem Darwishet al. and Younes Samih. 2021. [Pre-training bert on arabic tweets: Practical considerations](https://md.research.elamur.ai/free/en/papers/abdelali-2021-pre-training-bert-arabic-tweets/). Preprint, arXiv:2102.10684. [↗](https://arxiv.org/abs/2102.10684)
2. Wissam Antoun, Fady Balyet al. and Hazem Hajj. 2020. [AraBERT: Transformer-based model for Arabic language understanding](https://md.research.elamur.ai/free/en/papers/antoun-2020-arabert-transformer-based-model-arabic/). In Proceedings of the 4th Workshop on Open-Source Arabic Corpora and Processing Tools, with a Shared Task on Offensive Language Detection, pages 9–15, Marseille, France. European Language Resource Association.
3. Ting Wai Terence Au, Ingemar J. Coxet al. and Vasileios Lampos. 2022. E-ner – an annotated named entity recognition corpus of legal text. Preprint, arXiv:2212.09306. [↗](https://arxiv.org/abs/2212.09306)
4. BigScience Workshop et al. 2023. Bloom: A 176bparameter open-access multilingual language model. Preprint, arXiv:2211.05100. [↗](https://arxiv.org/abs/2211.05100)
5. Niama El Elkhbir, Urchade Zaratiana, Nadi Tomehet al. and Thierry Charnois. 2023. LIPN at Wojood-NER shared task: A span-based approach for flat and nested Arabic named entity recognition. In Proceedings of ArabicNLP 2023, pages 789–796, Singapore (Hybrid). Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Niama+El+Elkhbir%2C+Urchade+Zaratiana%2C+Nadi+Tomehet+al%2E+and+Thierry+Charnois%2E+2023%2E+LIPN+at+Wojood%2DNER+shared+task%3A+A+span%2Dbased+approach+for+flat+and+nested+Arabic+named+entity+recognition%2E+In+Proceedi)
6. Abdelkader El Mahdaouy, Salima Lamsiyah, Hamza Alami, Christoph Schommeret al. and Ismail Berrada. 2023. [UM6P & UL at WojoodNER shared task: Improving multi-task learning for flat and nested Arabic named entity recognition](https://md.research.elamur.ai/free/en/papers/mahdaouy-2023-um6p-ul-wojoodner-shared-task/). In Proceedings of ArabicNLP 2023, pages 777–782, Singapore (Hybrid). Association for Computational Linguistics. recognition shared task. In Proceedings of Arabic-NLP 2023, pages 748–758, Singapore (Hybrid). Association for Computational Linguistics.
7. Mustafa Jarrar, Nagham Hamad, Mohammed Khalilia, Bashar Talafhaet al. and Muhammad Elmadany, AbdelRahim Abdul-Mageed. 2024. [WojoodNER 2024: The Second Arabic Named Entity Recognition Shared Task](https://md.research.elamur.ai/free/en/papers/jarrar-2024-wojoodner-2024-second-arabic-named/). In Proceedings of the 2nd Arabic Natural Language Processing Conference (Arabic-NLP), Part of the ACL 2024. Association for Computational Linguistics.
8. Mustafa Jarrar, Mohammed Khaliliaet al. and Sana Ghanem. 2022. [Wojood: Nested Arabic named entity corpus and recognition using BERT](https://md.research.elamur.ai/free/en/papers/jarrar-2022-wojood-nested-arabic-named-entity/). In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 3626–3636, Marseille, France. European Language Resources Association.
9. Imen Laouirine, Haroun Elleuchet al. and Fethi Bougares. 2023. [ELYADATA at WojoodNER shared task: Data and model-centric approaches for Arabic flat and nested NER](https://md.research.elamur.ai/free/en/papers/laouirine-2023-elyadata-wojoodner-shared-task-data/). In Proceedings of ArabicNLP 2023, pages 759–764, Singapore (Hybrid). Association for Computational Linguistics.
10. Haneen Liqreina, Mustafa Jarrar, Mohammed Khalilia, Ahmed El-Shangitiet al. and Muhammad Abdul Mageed. 2023. [Arabic fine-grained entity recognition](https://md.research.elamur.ai/free/en/papers/liqreina-2023-arabic-fine-grained-entity-recognition/). In Proceedings of ArabicNLP 2023, pages 310–323.
11. Seungwhan Moon, Leonardo Neveset al. and Vitor Carvalho. 2018. Multimodal named entity recognition for short social media posts. CoRR, abs/1802.07862. [↗](https://arxiv.org/abs/1802.07862)
12. Gemma Team et al. 2024. Gemma: Open models based on gemini research and technology. Preprint, arXiv:2403.08295. [↗](https://arxiv.org/abs/2403.08295)
13. Niklas Muennighoff, Thomas Wang, Lintang Sutawika, Adam Roberts, Stella Biderman, Teven Le Scao, M Saiful Bari, Sheng Shen, Zheng Xin Yong, Hailey Schoelkopf, Xiangru Tang, Dragomir Radev, Alham Fikri Aji, Khalid Almubarak, Samuel Albanie, Zaid Alyafeai, Albert Webson, Edward Raffet al. and Colin Raffel. 2023. [Crosslingual generalization through multitask finetuning](https://md.research.elamur.ai/free/en/papers/muennighoff-2022-crosslingual-generalization-through-multitask-finetuning/). In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 15991–16111, Toronto, Canada. Association for Computational Linguistics.
14. Abbas Ghaddar, Yimeng Wu, Ahmad Rashid, Khalil Bibi, Mehdi Rezagholizadeh, Chao Xing, Yasheng Wang, Duan Xinyu, Zhefeng Wang, Baoxing Huai, Xin Jiang, Qun Liuet al. and Philippe Langlais. 2022. [Jaber and saber: Junior and senior arabic bert](https://md.research.elamur.ai/free/en/papers/ghaddar-2021-jaber-saber-junior-senior-arabic/). Preprint, arXiv:2112.04329. [↗](https://arxiv.org/abs/2112.04329)
15. Adrien Pavao, Isabelle Guyon, Anne-Catherine Letournel, Dinh-Tuan Tran, Xavier Baro, Hugo Jair Escalante, Sergio Escalera, Tyler Thomaset al. and Zhen Xu. 2023. Codalab competitions: An open source platform to organize scientific challenges. Journal of Machine Learning Research, 24(198):1–6. [find](https://scholar.google.com/scholar?q=Adrien+Pavao%2C+Isabelle+Guyon%2C+Anne%2DCatherine+Letournel%2C+Dinh%2DTuan+Tran%2C+Xavier+Baro%2C+Hugo+Jair+Escalante%2C+Sergio+Escalera%2C+Tyler+Thomaset+al%2E+and+Zhen+Xu%2E+2023%2E+Codalab+competitions%3A+An+open+source+pl)
16. Go Inoue, Bashar Alhafni, Nurpeiis Baimukan, Houda Bouamoret al. and Nizar Habash. 2021. [The interplay of variant, size, and task type in arabic pre-trained language models](https://md.research.elamur.ai/free/en/papers/inoue-2021-interplay-variant-size-task-type/). Preprint, arXiv:2103.06678. [↗](https://arxiv.org/abs/2103.06678)
17. Nadeesha Perera, Matthias Dehmeret al. and Frank Emmert-Streib. 2020. Named entity recognition and relation detection for biomedical information extraction. Frontiers in cell and developmental biology, 8:673. [find](https://scholar.google.com/scholar?q=Nadeesha+Perera%2C+Matthias+Dehmeret+al%2E+and+Frank+Emmert%2DStreib%2E+2020%2E+Named+entity+recognition+and+relation+detection+for+biomedical+information+extraction%2E+Frontiers+in+cell+and+developmental+biology)
18. Shereen Elkordi, Noha Adlyet al. and Marwan Torki. 2023. [AlexU-AIC at WojoodNER shared task: Sequence labeling vs MRC and SWA for Arabic named entity recognition](https://md.research.elamur.ai/free/en/papers/elkordi-2023-alexu-aic-wojoodner-shared-task/). In Proceedings of ArabicNLP 2023, pages 771–776, Singapore (Hybrid). Association for Computational Linguistics.
19. Mustafa Jarrar, Muhammad Abdul-Mageed, Mohammed Khalilia, Bashar Talafha, AbdelRahim Elmadany, Nagham Hamadet al. and Alaa’ Omar. 2023. WojoodNER 2023: The first Arabic named entity Russell Richie, Victor M Ruiz, Sifei Han, Lingyun Shiet al. and Fuchiang (Rich) Tsui. 2023. Extracting social determinants of health events with transformer-based multitask, multilabel named entity recognition. Journal of the American Medical Informatics Association, 30(8):1379–1388. [find](https://scholar.google.com/scholar?q=Mustafa+Jarrar%2C+Muhammad+Abdul%2DMageed%2C+Mohammed+Khalilia%2C+Bashar+Talafha%2C+AbdelRahim+Elmadany%2C+Nagham+Hamadet+al%2E+and+Alaa%E2%80%99+Omar%2E+2023%2E+WojoodNER+2023%3A+The+first+Arabic+named+entity+Russell+Richie%2C+Vi)

*Updated 2 November 2026*