---
title: "rematchka at ArAIEval Shared Task: Prefix-Tuning &"
description: "The paper describes the \"rematchka\" entry to the ArAIEval shared task at ArabicNLP 2023. 23 references, cited by 1 paper here."
authors: ["Reem Abdel-Salam"]
year: 2023
venue: "Proceedings of ArabicNLP 2023"
source: "https://aclanthology.org/2023.arabicnlp-1.52/"
doi: "10.18653/v1/2023.arabicnlp-1.52"
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/salam-2023-rematchka-araieval-shared-task-prefix/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/salam-2023-rematchka-araieval-shared-task-prefix/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2023](https://md.research.elamur.ai/free/en/papers/2023/page/13/) / rematchka at ArAIEval Shared Task: Prefix-Tuning & Prompt-tuning for Improved Detection of Propaganda and Disinformation in Arabic Social Media Content

# rematchka at ArAIEval Shared Task: Prefix-Tuning & Prompt-tuning for Improved Detection of Propaganda and Disinformation in Arabic Social Media Content

**Authors:** [Reem Abdel-Salam](https://md.research.elamur.ai/free/en/authors/reem-abdel-salam/) · **Year:** 2023 · **Venue:** [Proceedings of ArabicNLP 2023](https://md.research.elamur.ai/free/en/venues/arabicnlp/) ([ArabicNLP 2023](https://md.research.elamur.ai/free/en/venues/arabicnlp/2023/)) · **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/2023.arabicnlp-1.52/) · [PDF](https://aclanthology.org/2023.arabicnlp-1.52.pdf) · [DOI 10.18653/v1/2023.arabicnlp-1.52](https://doi.org/10.18653/v1/2023.arabicnlp-1.52) · [OpenAlex](https://openalex.org/W4389518447)

**Citations:** 2 citations (OpenAlex, 2026-10-14) · [1](https://md.research.elamur.ai/free/en/papers/salam-2023-rematchka-araieval-shared-task-prefix/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper describes the "rematchka" entry to the ArAIEval shared task at ArabicNLP 2023. The task covers two problems in Arabic tweets and texts. Task 1 asks whether a text uses persuasion (1A) and, if so, which of 24 propaganda techniques appear (1B, multi-label). Task 2 asks whether a text is disinformation (2A) and, if it is, which kind (2B). The organisers' data is heavily skewed toward some classes.

The author built weighted ensembles of AraBERT v2, MARBERT and AraGPT models and trained them several ways:
- conventional fine-tuning;
- parameter-efficient prefix-tuning;
- prompt-tuning through OpenPrompt.

Several robustness and imbalance tricks were layered on top: FGM adversarial perturbation of embeddings, supervised contrastive loss, Arabic–English back-translation to upsample minority classes, imbalance-aware losses, and a balanced multi-label sampler.

The system placed third in 1A (micro-F1 0.7555), second in 1B (0.5658) and fourth in 2A (0.9040) and 2B (0.8219). The author's conclusion is that prompt- and prefix-based tuning beat plain fine-tuning, and that losses designed for class imbalance help on the skewed labels.

## Key points
- **Data sizes (train/dev/test):**
  - 1A and 1B: 2427/259/503.
  - 2A: 14147/2111/3729.
  - 2B: 2656/397/876.
- **Class imbalance:**
  - 2A has 2656 disinformation versus 11,491 non-disinformation examples.
  - 2B labels are hate speech (1512), offensive (500), spam (453) and rumor (191).
- **Subtask 1B:** all models used prefix-tuning. Asymmetric Loss and Distribution Balanced Loss, combined with a balanced sampler, beat binary cross-entropy with random sampling on dev (64.89 and 66 versus 62).
- **Subtask 2B:** prompt-tuning used an Arabic cloze template meaning "The disinformation class is [MASK]". The answer words were the label names translated into Arabic.
- **Dev versus test:** model rankings on dev frequently reversed on test. In 1A, a single MARBERT model with prefix-tuning plus contrastive loss reached 76.5 on test, which would have beaten the submitted ensemble (75.55).

## Interesting ideas
- Casting classification as a fill-in-the-mask sentence whose answer words are the label names reuses the language model's own head instead of a freshly initialised classifier, which suits small, skewed label sets.
- Combining an asymmetric, imbalance-aware loss with a sampler that rebalances multi-label batches lifted rare-class performance over plain cross-entropy with random batches.
- Adding a small gradient-direction perturbation to the input embeddings during training acts as a cheap regulariser that pushes near-identical inputs toward the same output.
- Models that ranked best on dev often slipped on test, so a single dev split is a poor basis for choosing a model.
- The error analysis separated two causes of failure: missing background knowledge (a COVID-era news item) and genre confusion (a poem read as news).

## For Mova

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

- **I1** (learn): A cloze-style LLM witness could vote on speech-act or irony labels for silver data, kept as one explainable vote among several witnesses rather than the final decision.
- **I2** (tree): gparse could rebalance batches and weight losses for rare deprels (e.g., vocative, dislocated) that the treebanks seldom contain.
- **I3** (tree): Applying FGM to gparse's embedding layer may make parses more stable under spelling variants and noisy tokens.
- **I4** (learn): Evaluation should report rankings across several held-out splits and negative controls before a training trick is adopted.
- **I5** (world): The text world could record a genre/register attribute and flag facts that depend on outside knowledge, so readers know when a judgment needs knowledge-graph support.

*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

- [Cairo University](https://md.research.elamur.ai/free/en/universities/cairo-university/) — University, Giza, Egypt · 42 works · 1,350 citations

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,772 works · [🗄️ Information Systems](https://md.research.elamur.ai/free/en/shelves/computer-science/information-systems/) · 1,437 works · [🤝 Sociology and Political Science](https://md.research.elamur.ai/free/en/shelves/social-sciences/sociology-and-political-science/) · 577 works

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Reem Abdel-Salam. 2023. rematchka at ArAIEval Shared Task: Prefix-Tuning & Prompt-tuning for Improved Detection of Propaganda and Disinformation in Arabic Social Media Content. Proceedings of ArabicNLP 2023.

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

```bibtex
@inproceedings{abdelsalam2023rematchka,
  title = {{rematchka at ArAIEval Shared Task: Prefix-Tuning \& Prompt-tuning for Improved Detection of Propaganda and Disinformation in Arabic Social Media Content}},
  author = {Reem Abdel-Salam},
  booktitle = {Proceedings of ArabicNLP 2023},
  year = {2023},
  doi = {10.18653/v1/2023.arabicnlp-1.52},
  url = {https://doi.org/10.18653/v1/2023.arabicnlp-1.52}
}
```

- [← previous PROPRES: Investigating the Projectivity of Presupposition with Various Triggers and Environments](https://md.research.elamur.ai/free/en/papers/asami-2023-propres-investigating-projectivity-presupposition-various/)
- [next → rematchka at NADI 2023 shared task: Parameter Efficient tuning for Dialect Identification and Dialect Machine Translation](https://md.research.elamur.ai/free/en/papers/salam-2023-rematchka-nadi-2023-shared-task/)
- [↑ back to papers of 2023](https://md.research.elamur.ai/free/en/papers/2023/page/13/)

## References

23 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Muhammad Abdul-Mageed, AbdelRahim Elmadany, and El Moatez Billah Nagoudi. 2020. [Arbert & marbert: deep bidirectional transformers for arabic](https://md.research.elamur.ai/free/en/papers/abdul-mageed-2021-arbert-marbert-deep-bidirectional-transformers/). arXiv preprint arXiv:2101.01785. [↗](https://arxiv.org/abs/2101.01785)
2. Firoj Alam, Hamdy Mubarak, Wajdi Zaghouani, Giovanni Da San Martino, and Preslav Nakov. 2022. [Overview of the WANLP 2022 shared task on propaganda detection in Arabic](https://md.research.elamur.ai/free/en/papers/alam-2022-overview-wanlp-2022-shared-task/). In Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP), pages 108–118, Abu Dhabi, United Arab Emirates (Hybrid). Association for Computational Linguistics.
3. Wissam Antoun, Fady Baly, 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/). arXiv preprint arXiv:2003.00104. [↗](https://arxiv.org/abs/2003.00104)
4. Joseph Attieh and Fadi Hassan. 2022. [Pythoneers at wanlp 2022 shared task: Monolingual arabert for arabic propaganda detection and span extraction](https://md.research.elamur.ai/free/en/papers/attieh-2022-pythoneers-wanlp-2022-shared-task/). In Proceedings of the The Seventh Arabic Natural Language Processing Workshop (WANLP), pages 534– 540.
5. Qianben Chen, Richong Zhang, Yaowei Zheng, and Yongyi Mao. 2022. Dual contrastive learning: Text classification via label-aware data augmentation. arXiv preprint arXiv:2201.08702. [↗](https://arxiv.org/abs/2201.08702)
6. Jianfeng Chi, William Shand, Yaodong Yu, Kai-Wei Chang, Han Zhao, and Yuan Tian. 2022. Conditional supervised contrastive learning for fair text classification. arXiv preprint arXiv:2205.11485. [↗](https://arxiv.org/abs/2205.11485)
7. Ning Ding, Shengding Hu, Weilin Zhao, Yulin Chen, Zhiyuan Liu, Hai-Tao Zheng, and Maosong Sun. 2021. [Openprompt: An open-source framework for prompt-learning](https://md.research.elamur.ai/free/en/papers/ding-2022-openprompt-open-source-framework-prompt/). arXiv preprint arXiv:2111.01998. [↗](https://arxiv.org/abs/2111.01998)
8. Zihao Fu, Haoran Yang, Anthony Man-Cho So, Wai Lam, Lidong Bing, and Nigel Collier. 2023. On the effectiveness of parameter-efficient fine-tuning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 12799–12807. [find](https://scholar.google.com/scholar?q=Zihao+Fu%2C+Haoran+Yang%2C+Anthony+Man%2DCho+So%2C+Wai+Lam%2C+Lidong+Bing%2C+and+Nigel+Collier%2E+2023%2E+On+the+effectiveness+of+parameter%2Defficient+fine%2Dtuning%2E+In+Proceedings+of+the+AAAI+Conference+on+Artificial+I)
9. Ian J Goodfellow, Jonathon Shlens, and Christian Szegedy. 2014. [Explaining and harnessing adversarial examples](https://md.research.elamur.ai/free/en/papers/goodfellow-2014-explaining-harnessing-adversarial-examples/). arXiv preprint arXiv:1412.6572. [↗](https://arxiv.org/abs/1412.6572)
10. Maram Hasanain, Firoj Alam, Hamdy Mubarak, Samir Abdaljalil, Wajdi Zaghouani, Preslav Nakov, Giovanni Da San Martino, and Abed Alhakim Freihat. 2023. [ArAIEval Shared Task: Persuasion Techniques and Disinformation Detection in Arabic Text](https://md.research.elamur.ai/free/en/papers/hasanain-2023-araieval-shared-task-persuasion-techniques/). In Proceedings of the First Arabic Natural Language Processing Conference (ArabicNLP 2023), Singapore. Association for Computational Linguistics.
11. Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021a. [Lora: Low-rank adaptation of large language models](https://md.research.elamur.ai/free/en/papers/hu-2021-lora-low-rank-adaptation-large/). arXiv preprint arXiv:2106.09685. [↗](https://arxiv.org/abs/2106.09685)
12. Shengding Hu, Ning Ding, Huadong Wang, Zhiyuan Liu, Jingang Wang, Juanzi Li, Wei Wu, and Maosong Sun. 2021b. [Knowledgeable prompt-tuning: Incorporating knowledge into prompt verbalizer for text classification](https://md.research.elamur.ai/free/en/papers/hu-2022-knowledgeable-prompt-tuning-incorporating-knowledge/). arXiv preprint arXiv:2108.02035. [↗](https://arxiv.org/abs/2108.02035)
13. Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan. 2020. Supervised contrastive learning. Advances in neural information processing systems, 33:18661–18673. [↗](https://doi.org/10.48550/arxiv.2004.11362)
14. Sahinur Rahman Laskar, Rahul Singh, Abdullah Faiz Ur Rahman Khilji, Riyanka Manna, Partha Pakray, and Sivaji Bandyopadhyay. 2022. [Cnlp-nits-pp at wanlp 2022 shared task: Propaganda detection in arabic using data augmentation and arabert pre-trained model](https://md.research.elamur.ai/free/en/papers/laskar-2022-cnlp-nits-pp-wanlp-2022/). In Proceedings of the The Seventh Arabic Natural Language Processing Workshop (WANLP), pages 541–544.
15. Brian Lester, Rami Al-Rfou, and Noah Constant. 2021. [The power of scale for parameter-efficient prompt tuning](https://md.research.elamur.ai/free/en/papers/lester-2021-power-scale-parameter-efficient-prompt/). arXiv preprint arXiv:2104.08691. [↗](https://arxiv.org/abs/2104.08691)
16. Xiang Lisa Li and Percy Liang. 2021. Prefix-tuning: Optimizing continuous prompts for generation. arXiv preprint arXiv:2101.00190. [↗](https://arxiv.org/abs/2101.00190)
17. Haokun Liu, Derek Tam, Mohammed Muqeeth, Jay Mohta, Tenghao Huang, Mohit Bansal, and Colin A Raffel. 2022. [Few-shot parameter-efficient fine-tuning is better and cheaper than in-context learning](https://md.research.elamur.ai/free/en/papers/liu-2022-few-shot-parameter-efficient-fine/). Advances in Neural Information Processing Systems, 35:1950–1965.
18. Lin Pan, Chung-Wei Hang, Avirup Sil, and Saloni Potdar. 2022. Improved text classification via contrastive adversarial training. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 11130–11138. [find](https://scholar.google.com/scholar?q=Lin+Pan%2C+Chung%2DWei+Hang%2C+Avirup+Sil%2C+and+Saloni+Potdar%2E+2022%2E+Improved+text+classification+via+contrastive+adversarial+training%2E+In+Proceedings+of+the+AAAI+Conference+on+Artificial+Intelligence%2C+volum)
19. Tal Ridnik, Emanuel Ben-Baruch, Nadav Zamir, Asaf Noy, Itamar Friedman, Matan Protter, and Lihi Zelnik-Manor. 2021. Asymmetric loss for multi-label classification. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 82–91. [↗](https://doi.org/10.48550/arxiv.2009.14119)
20. Ahmed Samir, Abu Bakr Soliman, Mohamed Ibrahim, Laila Hesham, and Samhaa R El-Beltagy. 2022. Ngu_cnlp at wanlp 2022 shared task: Propaganda detection in arabic. WANLP 2022, page 545. [find](https://scholar.google.com/scholar?q=Ahmed+Samir%2C+Abu+Bakr+Soliman%2C+Mohamed+Ibrahim%2C+Laila+Hesham%2C+and+Samhaa+R+El%2DBeltagy%2E+2022%2E+Ngu%5Fcnlp+at+wanlp+2022+shared+task%3A+Propaganda+detection+in+arabic%2E+WANLP+2022%2C+page+545%2E)
21. Xiaosen Wang, Yichen Yang, Yihe Deng, and Kun He. 2021. Adversarial training with fast gradient projection method against synonym substitution based text attacks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 13997– 14005. [find](https://scholar.google.com/scholar?q=Xiaosen+Wang%2C+Yichen+Yang%2C+Yihe+Deng%2C+and+Kun+He%2E+2021%2E+Adversarial+training+with+fast+gradient+projection+method+against+synonym+substitution+based+text+attacks%2E+In+Proceedings+of+the+AAAI+Conference)
22. Yaqing Wang, Subhabrata Mukherjee, Xiaodong Liu, Jing Gao, Ahmed Hassan Awadallah, and Jianfeng Gao. 2022. Adamix: Mixture-of-adapter for parameter-efficient tuning of large language models. arXiv preprint arXiv:2205.12410, 1(2):4. [↗](https://arxiv.org/abs/2205.12410)
23. Tong Wu, Qingqiu Huang, Ziwei Liu, Yu Wang, and Dahua Lin. 2020. Distribution-balanced loss for multi-label classification in long-tailed datasets. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part IV 16, pages 162–178. Springer. [find](https://scholar.google.com/scholar?q=Tong+Wu%2C+Qingqiu+Huang%2C+Ziwei+Liu%2C+Yu+Wang%2C+and+Dahua+Lin%2E+2020%2E+Distribution%2Dbalanced+loss+for+multi%2Dlabel+classification+in+long%2Dtailed+datasets%2E+In+Computer+Vision%E2%80%93ECCV+2020%3A+16th+European+Conferen)

*Updated 1 November 2026*