---
title: "MAPROC at AHaSIS Shared Task — paper by Zarnoufi (2025)"
description: "The paper describes a system for the AHaSIS shared task. 78.87% F1. 22 references. Among them: 0."
authors: ["Randa Zarnoufi"]
year: 2025
venue: "Proceedings of the Shared Task on Sentiment Analysis for Arabic Dialects"
source: "https://aclanthology.org/2025.ranlp-ahasis.8/"
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/zarnoufi-2025-maproc-ahasis-shared-task-few/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/zarnoufi-2025-maproc-ahasis-shared-task-few/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2025](https://md.research.elamur.ai/free/en/papers/2025/page/18/) / MAPROC at AHaSIS Shared Task: Few-Shot and Sentence Transformer for Sentiment Analysis of Arabic Hotel Reviews

# MAPROC at AHaSIS Shared Task: Few-Shot and Sentence Transformer for Sentiment Analysis of Arabic Hotel Reviews

**Authors:** [Randa Zarnoufi](https://md.research.elamur.ai/free/en/authors/randa-zarnoufi/) · **Year:** 2025 · **Venue:** [Proceedings of the Shared Task on Sentiment Analysis for Arabic Dialects](https://md.research.elamur.ai/free/en/venues/ranlp/) ([RANLP 2025](https://md.research.elamur.ai/free/en/venues/ranlp/2025/)) · **Publisher:** [INCOMA Ltd.](https://md.research.elamur.ai/free/en/publishers/incoma-ltd/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2025.ranlp-ahasis.8/) · [PDF](https://aclanthology.org/2025.ranlp-ahasis.8.pdf) · [OpenAlex](https://openalex.org/W7106286089)

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

## Summary
The paper describes a system for the AHaSIS shared task. The task asks for three-way sentiment labels (positive, neutral, negative) on hotel reviews written in two Arabic dialects, Moroccan Darija and Saudi. Annotated dialectal data is scarce, so the author does not rely on full fine-tuning of a large model. Instead the system uses SetFit, a few-shot method with two stages. First, a sentence transformer (Arabic-SBERT-100K) is adapted with a contrastive objective on pairs built from the labelled reviews. Second, a lightweight classification head is trained on the resulting embeddings. Preprocessing is light: punctuation is removed and alif/hamza spellings are normalized. With only 64 examples per class, the system scored 78.87% F1 on an internal held-out split and 73% F1 on the official test set, placing 12th of 26 teams; the winning team reached 81%. The work shows that a cheap, quick-to-train few-shot pipeline can reach competitive results for under-resourced dialects in a narrow domain. It also shows that the choice of base encoder has a large effect.

## Key points
- Data: 860 training reviews (430 Darija, 430 Saudi) and 216 evaluation reviews (108 per dialect). The class split is about 39% negative, 36% positive and 25% neutral.
- Method: SetFit with Arabic-SBERT-100K as the body. This is an SBERT variant built on AraBERT and tuned on 100K Arabic sentences. The comparison baseline is AraBERTv0.2 base fine-tuned on the full training set, and the internal split is 80/20.
- Data efficiency: F1 rose steadily as the number of examples per class grew from 8 to 64. The best internal score was 78.87% F1, obtained from just 192 labelled reviews in total.
- Encoder choice: swapping in ArabicBERT_Finetuned-AR-500 as the sentence encoder dropped F1 to 52.83% (3 epochs).
- Per-dialect analysis: Saudi exceeded 80% F1 in every class, including neutral (80%). Darija was weaker, especially on neutral reviews (61.3%).

## Interesting ideas
- Turning a small labelled set into many same-label and different-label pairs multiplies the training signal, so an encoder can be adapted from only dozens of examples per class.
- The representation is tuned separately from the final decision, which is a plain head such as logistic regression, so the last step stays simple and easy to inspect.
- Breaking scores down by dialect and by class exposed a specific weak cell, Darija neutral, that the overall F1 concealed.
- Measuring performance at 8, 16, 32 and 64 examples per class gives a learning curve that tells you how much annotation is still worth paying for.
- With the method held fixed, changing only the base encoder moved F1 by more than 20 points, so the foundation model should be chosen as its own experiment.

## For Mova

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

- **I1** (frames): A small set of hand-checked speech-act or irony examples could be expanded into pairs to adapt a compact encoder. That encoder could then act as an extra, cheap witness next to the rule-based prag decisions.
- **I2** (learn): Where Mova needs a learned scorer, it could keep the model's final layer as a transparent linear head over the model's embeddings. Its weights and the influence of each example would then remain auditable, in keeping with the explainability requirement.
- **I3** (learn): Evaluation reports should break results down by genre (fables, word problems, school tests) and by label type. That would surface narrow failure pockets that a single aggregate number hides.
- **I4** (learn): Plotting such curves for the LLM teacher tiers would show at what point more teacher-labelled silver data stops helping a given component.
- **I5** (tree): For gparse, several candidate pretrained encoders should be benchmarked under an identical training recipe before tuning anything else.

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

📚 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 (2): [😊 Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/) · 1,035 works · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 3,066 works

## Cite

Cite as: Randa Zarnoufi. 2025. MAPROC at AHaSIS Shared Task: Few-Shot and Sentence Transformer for Sentiment Analysis of Arabic Hotel Reviews. Proceedings of the Shared Task on Sentiment Analysis for Arabic Dialects.

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

```bibtex
@inproceedings{zarnoufi2025maproc,
  title = {{MAPROC at AHaSIS Shared Task: Few-Shot and Sentence Transformer for Sentiment Analysis of Arabic Hotel Reviews}},
  author = {Randa Zarnoufi},
  booktitle = {Proceedings of the Shared Task on Sentiment Analysis for Arabic Dialects},
  year = {2025},
  url = {https://aclanthology.org/2025.ranlp-ahasis.8/}
}
```

- [← previous Lucas, Peter J. 2024. Old English Poetry from Manuscript to Message. Utrecht Studies in Medieval Literacy 58. Turnhout: Brepols. Pp. xviii + 398. ISBN 9782503600314.](https://md.research.elamur.ai/free/en/papers/griffith-2025-lucas-peter-j-2024-old/)
- [next → MarsadLab at NADI Shared Task: Arabic Dialect Identification and Speech Recognition using ECAPA-TDNN and Whisper](https://md.research.elamur.ai/free/en/papers/biswas-2025-marsadlab-nadi-shared-task-arabic/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/page/18/)

## References

22 references, 15 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Abdul-Mageed, M., Elmadany, A., Moatez, E., & Nagoudi, B. (2021). [ARBERT & MARBERT: Deep Bidirectional Transformers for Arabic](https://md.research.elamur.ai/free/en/papers/abdul-mageed-2021-arbert-marbert-deep-bidirectional-transformers/). [https://github.com/attardi/wikiextractor](https://github.com/attardi/wikiextractor). [↗](https://github.com/attardi/wikiextractor)
2. Aly, M., & Atiya, A. (2013). [LABR: A Large Scale Arabic Book Reviews Dataset](https://md.research.elamur.ai/free/en/papers/aly-2013-labr-large-scale-arabic-book/). In H. Schuetze, P. Fung, & M. Poesio (Eds.), Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) (pp. 494–498). Association for Computational Linguistics. https://aclanthology.org/P13-2088/ [↗](https://aclanthology.org/P13-2088/)
3. Alharbi, M., Chafik, S., Ezzini, S., Mitkov, R., Ranasinghe, T., & Hettiarachchi, H. (2025). [AHaSIS: Shared Task on Sentiment Analysis for Arabic Dialects](https://md.research.elamur.ai/free/en/papers/alharbi-2025-ahasis-shared-task-sentiment-analysis/). Proceedings of the 15th International Conference on Recent Advances in Natural Language Processing (RANLP).
4. Antoun, W., Baly, F., & Hajj, H. (2020). [AraBERT: Transformer-based Model for Arabic Language Understanding](https://md.research.elamur.ai/free/en/papers/antoun-2020-arabert-transformer-based-model-arabic/). http://arxiv.org/abs/2003.00104 [↗](https://arxiv.org/abs/2003.00104)
5. Alharbi, M., Ezzini, S., Hettiarachchi, H., Ranasinghe, T., & Mitkov, R. (2025). Evaluating Large Language Models on Arabic Dialect Sentiment Analysis. Proceedings of the 15th International Conference on Recent Advances in Natural Language Processing (RANLP). [find](https://scholar.google.com/scholar?q=Alharbi%2C+M%2E%2C+Ezzini%2C+S%2E%2C+Hettiarachchi%2C+H%2E%2C+Ranasinghe%2C+T%2E%2C+%26+Mitkov%2C+R%2E+%282025%29%2E+Evaluating+Large+Language+Models+on+Arabic+Dialect+Sentiment+Analysis%2E+Proceedings+of+the+15th+International+Conference)
6. Ashi, M. M., Siddiqui, M. A., & Nadeem, F. (2019). Pre-trained Word Embeddings for Arabic Aspect-Based Sentiment Analysis of Airline Tweets (A. E. Hassanien, M. F. Tolba, K. Shaalan, & A. T. Azar, Eds.; Vol. 845). Springer International Publishing. https://doi.org/10.1007/978-3-319-99010-1 [↗](https://doi.org/10.1007/978-3-319-99010-1)
7. Al-Horaibi, L., & Khan, M. B. (2016). Sentiment analysis of Arabic tweets using text mining Badaro, G., Baly, R., Hajj, H., Habash, N., & El-Hajj, W. (2014). [A Large Scale Arabic Sentiment Lexicon for Arabic Opinion Mining](https://md.research.elamur.ai/free/en/papers/badaro-2014-large-scale-arabic-sentiment-lexicon/). In N. Habash & S. Vogel (Eds.), Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP) (pp. 165–173). Association for Computational Linguistics. https://doi.org/10.3115/v1/W14-3623 [↗](https://doi.org/10.3115/v1/w14-3623)
8. Mulki, H., Haddad, * -Hatem, & Babao˘, I. (2017). [Modern Trends in Arabic Sentiment Analysis: A Survey](https://md.research.elamur.ai/free/en/papers/mulki-2017-modern-trends-arabic-sentiment-analysis/) (Vol. 58).
9. Nabil, M., Aly, M., & Atiya, A. (2015). [ASTD: Arabic Sentiment Tweets Dataset](https://md.research.elamur.ai/free/en/papers/nabil-2015-astd-arabic-sentiment-tweets-dataset/). In L. Màrquez, C. Callison-Burch, & J. Su (Eds.), Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing (pp. 2515–2519). Association for Computational Linguistics. https://doi.org/10.18653/v1/D15-1299 [↗](https://doi.org/10.18653/v1/d15-1299)
10. Baly, R., Khaddaj, A., Hajj, H. M., El-Hajj, W., & Shaban, K. B. (2019). [ArSentD-LEV: A Multi-Topic Corpus for Target-based Sentiment Analysis in Arabic Levantine Tweets](https://md.research.elamur.ai/free/en/papers/baly-2019-arsentd-lev-multi-topic-corpus/). CoRR, abs/1906.01830. http://arxiv.org/abs/1906.01830 [↗](https://arxiv.org/abs/1906.01830)
11. Salameh, M., Mohammad, S. M., & Kiritchenko, S. (2015). [Sentiment after Translation: A Case-Study on Arabic Social Media Posts](https://md.research.elamur.ai/free/en/papers/salameh-2015-sentiment-after-translation-case-study/). [http://www.purl.com/net/ArabicSentiment](http://www.purl.com/net/ArabicSentiment) [↗](http://www.purl.com/net/ArabicSentiment)
12. Dahou, A., Xiong, S., Zhou, J., Haddoud, M. H., & Duan, P. (2016). [Word Embeddings and Convolutional Neural Network for Arabic Sentiment Classification](https://md.research.elamur.ai/free/en/papers/dahou-2016-word-embeddings-convolutional-neural-network/). In Y. Matsumoto & R. Prasad (Eds.), Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers (pp. 2418–2427).
13. The COLING 2016 [Organizing Committee](https://md.research.elamur.ai/free/en/papers/dahou-2016-word-embeddings-convolutional-neural-network/). https://aclanthology.org/C16-1228/ [↗](https://aclanthology.org/C16-1228/)
14. Shi, Z., & Agrawal, R. (2025). [A comprehensive survey of contemporary Arabic sentiment analysis: Methods, Challenges, and Future Directions](https://md.research.elamur.ai/free/en/papers/shi-2025-comprehensive-survey-contemporary-arabic-sentiment/).
15. Soliman, A. B., Eissa, K., & El-Beltagy, S. R. (2017). AraVec: A set of Arabic Word Embedding Models for use in Arabic NLP. Procedia Computer Science, 117, 256–265. https://doi.org/10.1016/j.procs.2017.10.117 [↗](https://doi.org/10.1016/j.procs.2017.10.117)
16. Duwairi, R., Ahmed, N. A., & Al-Rifai, S. Y. (2015). Detecting sentiment embedded in Arabic social media - A lexicon-based approach. J. Intell. Fuzzy Syst., 29, 107–117. [https://api.semanticscholar.org/CorpusID:61](https://api.semanticscholar.org/CorpusID:61) 45765 [↗](https://api.semanticscholar.org/CorpusID:61)
17. Tunstall, L., Reimers, N., Jo, U. E. S., Bates, L., Korat, D., Wasserblat, M., & Pereg, O. (2022). Efficient Few-Shot Learning Without Prompts. http://arxiv.org/abs/2209.11055 [↗](https://arxiv.org/abs/2209.11055)
18. Eljundi, O., Antoun, W., Nour, ), Droubi, E., Hajj, H., El-Hajj, W., & Shaban, K. (2019). [hULMonA ( ): The Universal Language Model in Arabic](https://md.research.elamur.ai/free/en/papers/eljundi-2019-hulmona-universal-language-model-arabic/).
19. Fang, Y., & Xu, C. (2024). ArSen-20: A New Benchmark for Arabic Sentiment Detection. 5th Workshop on African Natural Language Processing. [https://openreview.net/forum?id=GgsRUF5k](https://openreview.net/forum?id=GgsRUF5k) [↗](https://openreview.net/forum?id=GgsRUF5k)
20. Jt Inoue, G., Alhafni, B., Baimukan, N., Bouamor, H., & Habash, N. (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/). http://arxiv.org/abs/2103.06678 [↗](https://arxiv.org/abs/2103.06678)
21. Khosla, P., Teterwak, P., Wang, C., Sarna, A., Tian, Y., Isola, P., Maschinot, A., Liu, C., & Krishnan, D. (2020). Supervised Contrastive Learning. http://arxiv.org/abs/2004.11362 [↗](https://arxiv.org/abs/2004.11362)
22. Mashaabi, M. O., Al-Khalifa, S. Z., & Al-Khalifa, H. S. (2024). [A Survey of Large Language Models for Arabic Language and its Dialects](https://md.research.elamur.ai/free/en/papers/mashaabi-2024-survey-large-language-models-arabic/). [https://mistral.ai/news/mistral-saba](https://mistral.ai/news/mistral-saba) [↗](https://mistral.ai/news/mistral-saba)

*Updated 2 November 2026*