---
title: "Legend at ArAIEval Shared Task — paper by Ojo et al. (2023)"
description: "The authors (team \"Legend\") describe their submission to Task 1 of the ArAIEval shared task at… 20 references, cited by 9 papers here."
authors: ["Olumide Ojo", "Olaronke Adebanji", "Hiram Calvo", "Damian Dieke", "Olumuyiwa Ojo", "Seye Akinsanya", "Tolulope Abiola", "Anna Feldman"]
year: 2023
venue: "Proceedings of ArabicNLP 2023"
source: "https://aclanthology.org/2023.arabicnlp-1.61/"
doi: "10.18653/v1/2023.arabicnlp-1.61"
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/ojo-2023-legend-araieval-shared-task-persuasion/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/ojo-2023-legend-araieval-shared-task-persuasion/)

[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/11/) / Legend at ArAIEval Shared Task: Persuasion Technique Detection using a Language-Agnostic Text Representation Model

# Legend at ArAIEval Shared Task: Persuasion Technique Detection using a Language-Agnostic Text Representation Model

**Authors:** Olumide Ojo, Olaronke Adebanji, [Hiram Calvo](https://md.research.elamur.ai/free/en/authors/hiram-calvo/), Damian Dieke, Olumuyiwa Ojo, Seye Akinsanya, [Tolulope Abiola](https://md.research.elamur.ai/free/en/authors/tolulope-olalekan-abiola/) and [Anna Feldman](https://md.research.elamur.ai/free/en/authors/anna-feldman/) · **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-1.61/) · [PDF](https://aclanthology.org/2023.arabicnlp-1.61.pdf) · [DOI 10.18653/v1/2023.arabicnlp-1.61](https://doi.org/10.18653/v1/2023.arabicnlp-1.61) · [OpenAlex](https://openalex.org/W4389518443)

**Citations:** 1 citations (OpenAlex, 2026-10-14) · [9](https://md.research.elamur.ai/free/en/papers/ojo-2023-legend-araieval-shared-task-persuasion/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The authors (team "Legend") describe their submission to Task 1 of the ArAIEval shared task at ArabicNLP 2023, subtask A. The task asks whether a short Arabic snippet, either a tweet or a paragraph from a news article, uses any persuasion technique. This is a binary decision. The organizers supplied labelled training and test data, with far more persuasive than non-persuasive examples.

The team fine-tuned the multilingual encoder XLM-RoBERTa through the Hugging Face Transformers library. They added a dropout layer and a classification head on top, and weighted the cross-entropy loss by class to counter the label skew. Training used Adam, a stepwise learning-rate decay and early stopping. They report trying other language models first and keeping XLM-RoBERTa because it performed best.

On the official test set the system reached a micro-F1 of 0.64, compared with 0.48 for the baseline. The paper shows that a plain, well-tuned multilingual fine-tuning recipe gives a solid result on Arabic persuasion detection. It also gives a reproducible reference configuration for similar low-resource classification tasks.

## Key points
- Data: 2,427 training snippets and 503 test snippets labelled true or false. The training labels are skewed, with 1,918 true and 509 false.
- Model: XLM-RoBERTa with a dropout layer and an added classification layer, fine-tuned end to end using Hugging Face Transformers.
- Hyperparameters:
  - learning rate 5×10⁻⁵
  - batch size 16
  - 6 epochs
  - Adam optimizer
  - early stopping on cross-entropy loss
  - StepLR scheduler that multiplies the rate by 0.85 every 2 epochs
- Class weights were applied in the loss to offset the imbalance between persuasive and non-persuasive texts.
- Test results (micro-F1, the official metric): 0.6402 for XLM-RoBERTa versus 0.4771 for the baseline.

## Interesting ideas
- Weighting the loss by class frequency stops a fine-tuned classifier from settling on the majority label when one class outnumbers the other almost four to one.
- A mild step decay of the learning rate (×0.85 every two epochs), paired with early stopping on validation loss, is a cheap and stable way to fine-tune a large encoder on a small dataset.
- The task is framed as a coarse binary gate (does the text persuade at all?) rather than jumping straight to fine-grained technique labels, which keeps the decision simple when labelled data is scarce.
- With about 79% of training items positive, the 0.64 micro-F1 is only meaningful next to a reference point; the paper shows the organizers' baseline, and a trivial majority-label predictor would be another obvious one.

## For Mova

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

- **I1** (learn): rare UD relations or rare frame roles in silver data could get inverse-frequency loss weights, so the trained parser does not neglect them.
- **I2** (tree): the gparse transformer parser could use this schedule when adapting to a small treebank or a converted UD dialect, avoiding overfitting without heavy tuning.
- **I3** (frames): the prag crate could first flag a clause as rhetorical or manipulative, then run its detailed speech-act, implicature or irony analysis only on flagged clauses, with the flag itself giving an explainable reason.
- **I4** (learn): any skewed-label evaluation should report a majority-class negative control next to the model score, so apparent gains are not just label frequency.

*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

- [Instituto Politécnico Nacional](https://md.research.elamur.ai/free/en/universities/instituto-politecnico-nacional/) — University, Mexico City, Mexico · 18 works · 4,307 citations
- [Montclair State University](https://md.research.elamur.ai/free/en/universities/montclair-state-university/) — University, Montclair, United States · 5 works · 94 citations

Also on the byline: [Federal University Oye Ekiti](https://ror.org/02q5h6807) · [Lead City University](https://ror.org/043z5qa52) · [Caritas University](https://ror.org/03p823x35).

📚 Shelves (4): [🖥️ 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 · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

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

## Cite

Cite as: Olumide Ojo, Olaronke Adebanji, Hiram Calvo, Damian Dieke, Olumuyiwa Ojo, Seye Akinsanya, Tolulope Abiola and Anna Feldman. 2023. Legend at ArAIEval Shared Task: Persuasion Technique Detection using a Language-Agnostic Text Representation Model. Proceedings of ArabicNLP 2023.

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

```bibtex
@inproceedings{ojo2023legend,
  title = {{Legend at ArAIEval Shared Task: Persuasion Technique Detection using a Language-Agnostic Text Representation Model}},
  author = {Olumide Ojo, Olaronke Adebanji, Hiram Calvo, Damian Dieke, Olumuyiwa Ojo, Seye Akinsanya, Tolulope Abiola and Anna Feldman},
  booktitle = {Proceedings of ArabicNLP 2023},
  year = {2023},
  doi = {10.18653/v1/2023.arabicnlp-1.61},
  url = {https://doi.org/10.18653/v1/2023.arabicnlp-1.61}
}
```

- [← previous Learning and Leveraging Verifiers to Improve Planning Capabilities of Pre-trained Language Models](https://md.research.elamur.ai/free/en/papers/arora-2023-learning-leveraging-verifiers-improve-planning/)
- [next → Mapping Explicit and Implicit Discourse Relations between the RST-DT and the PDTB 3.0](https://md.research.elamur.ai/free/en/papers/costa-2023-mapping-explicit-implicit-discourse-relations/)
- [↑ back to papers of 2023](https://md.research.elamur.ai/free/en/papers/2023/page/11/)

## References

20 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Mohamed Abd Elaziz, Abdelghani Dahou, Dina Ahmed Orabi, Samah Alshathri, Eman M Soliman, and Ahmed A Ewees. 2023. A hybrid multitask learning framework with a fire hawk optimizer for arabic fake news detection. Mathematics, 11(2):258. [↗](https://doi.org/10.3390/math11020258)
2. Hu Huafeng, Fan Zhihao, and Gao Xuezhen. 2019. The history of the arab nation and the arabic language. Arabic Language, Literature & Culture, 4(3):48. [↗](https://doi.org/10.11648/j.allc.20190403.11)
3. Dia Abujaber, Ahmed Qarqaz, and Malak A Abdullah. 2021. Lecun at semeval-2021 task 6: detecting persuasion techniques in text using ensembled pretrained transformers and data augmentation. In Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 1068–1074. [find](https://scholar.google.com/scholar?q=Dia+Abujaber%2C+Ahmed+Qarqaz%2C+and+Malak+A+Abdullah%2E+2021%2E+Lecun+at+semeval%2D2021+task+6%3A+detecting+persuasion+techniques+in+text+using+ensembled+pretrained+transformers+and+data+augmentation%2E+In+Proceedi)
4. 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 The Seventh Arabic Natural Language Processing Workshop (WANLP), pages 108–118, Abu Dhabi, United Arab Emirates (Hybrid). Association for Computational Linguistics.
5. Ahmed Aleroud, Abdullah Bani Melhem, Nour Al-hussien, and Craig Douglas Albert. 2023. span-prop: Combatting contextualized social media state-linked propaganda in the middle east. [find](https://scholar.google.com/scholar?q=Ahmed+Aleroud%2C+Abdullah+Bani+Melhem%2C+Nour+Al%2Dhussien%2C+and+Craig+Douglas+Albert%2E+2023%2E+span%2Dprop%3A+Combatting+contextualized+social+media+state%2Dlinked+propaganda+in+the+middle+east%2E)
6. Marietjie Botes. 2023. Autonomy and the social dilemma of online manipulative behavior. AI and Ethics, 3(1):315–323. [↗](https://doi.org/10.1007/s43681-022-00157-5)
7. Nikhilesh Dholakia, Aras Ozgun, and Deniz Atik. 2023. The miasma of misinformation: a social analysis of media, markets, and manipulation. Consumption Markets & Culture, pages 1–16. [↗](https://doi.org/10.1080/10253866.2022.2149508)
8. Sabah Ghazzawi. 1992. The arabic language. Washington DC: Center for Contemporary Arab Studies. [find](https://scholar.google.com/scholar?q=Sabah+Ghazzawi%2E+1992%2E+The+arabic+language%2E+Washington+DC%3A+Center+for+Contemporary+Arab+Studies%2E)
9. Kshitij Gupta, Devansh Gautam, and Radhika Mamidi. 2021. Volta at semeval-2021 task 6: Towards detecting persuasive texts and images using textual and multimodal ensemble. arXiv preprint arXiv:2106.00240. [↗](https://arxiv.org/abs/2106.00240)
10. Hasanain Maram, El-Shangiti Ahmed Oumar, Nandi Rabindra Nath, Nakov Preslav, and Alam Firoj. 2023. Qcri at semeval-2023 task 3: News genre, framing and persuasion techniques detection using multilingual models. arXiv preprint arXiv:2305.03336. [↗](https://arxiv.org/abs/2305.03336)
11. Arkadiusz Modzelewski, Witold Sosnowski, Magdalena Wilczynska, and Adam Wierzbicki. 2023. Dshacker at semeval-2023 task 3: Genres and persuasion techniques detection with multilingual data augmentation through machine translation and text generation. In Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1582–1591. [find](https://scholar.google.com/scholar?q=Arkadiusz+Modzelewski%2C+Witold+Sosnowski%2C+Magdalena+Wilczynska%2C+and+Adam+Wierzbicki%2E+2023%2E+Dshacker+at+semeval%2D2023+task+3%3A+Genres+and+persuasion+techniques+detection+with+multilingual+data+augmentatio)
12. Nikolaos Nikolaidis, Nicolas Stefanovitch, and Jakub Piskorski. 2023. On experiments of detecting persuasion techniques in polish and russian online news: Preliminary study. In Proceedings of the 9th Workshop on Slavic Natural Language Processing 2023 (SlavicNLP 2023), pages 155–164. [find](https://scholar.google.com/scholar?q=Nikolaos+Nikolaidis%2C+Nicolas+Stefanovitch%2C+and+Jakub+Piskorski%2E+2023%2E+On+experiments+of+detecting+persuasion+techniques+in+polish+and+russian+online+news%3A+Preliminary+study%2E+In+Proceedings+of+the+9th+)
13. OE Ojo, A Gelbukh, H Calvo, and OO Adebanji. 2021. Performance study of n-grams in the analysis of sentiments. Journal of the Nigerian Society of Physical Sciences, pages 477–483. [↗](https://doi.org/10.46481/jnsps.2021.201)
14. OE Ojo, A Gelbukh, H Calvo, A Feldman, OO Adebanji, and J Armenta-Segura. 2022a. Language identification at the word level in code-mixed texts using character sequence and word embedding. In Proceedings of the 19th International Conference on Natural Language Processing (ICON): Shared Task on Word Level Language Identification in Code-mixed Kannada-English Texts, pages 1–6. [find](https://scholar.google.com/scholar?q=OE+Ojo%2C+A+Gelbukh%2C+H+Calvo%2C+A+Feldman%2C+OO+Adebanji%2C+and+J+Armenta%2DSegura%2E+2022a%2E+Language+identification+at+the+word+level+in+code%2Dmixed+texts+using+character+sequence+and+word+embedding%2E+In+Proceedin)
15. Olumide E Ojo, Alexander Gelbukh, Hiram Calvo, Olaronke O Adebanji, and Grigori Sidorov. 2020. Sentiment detection in economics texts. In Mexican International Conference on Artificial Intelligence, pages 271–281. Springer. [find](https://scholar.google.com/scholar?q=Olumide+E+Ojo%2C+Alexander+Gelbukh%2C+Hiram+Calvo%2C+Olaronke+O+Adebanji%2C+and+Grigori+Sidorov%2E+2020%2E+Sentiment+detection+in+economics+texts%2E+In+Mexican+International+Conference+on+Artificial+Intelligence%2C+p)
16. Olumide Ebenezer Ojo, Thang Ta Hoang, Alexander Gelbukh, Hiram Calvo, Grigori Sidorov, and Olaronke Oluwayemisi Adebanji. 2022b. Automatic hate speech detection using cnn model and word embedding. Computación y Sistemas, 26(2). [↗](https://doi.org/10.13053/cys-26-2-4107)
17. Olumide Ebenezer Ojo, Hoang Thang Ta, Alexander Gelbukh, Hiram Calvo, Olaronke Oluwayemisi Adebanji, and Grigori Sidorov. 2023. Transformer-based approaches to sentiment detection. In Recent Developments and the New Directions of Research, Foundations, and Applications: Selected Papers of the 8th World Conference on Soft Computing, February 03– 05, 2022, Baku, Azerbaijan, Vol. II, pages 101–110. Springer. [↗](https://doi.org/10.1007/978-3-031-23476-7_10)
18. Jakub Piskorski, Nicolas Stefanovitch, Giovanni Da San Martino, and Preslav Nakov. 2023. Semeval-2023 task 3: Detecting the category, the framing, and the persuasion techniques in online news in a multilingual setup. In Proceedings of the the 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 2343–2361. [find](https://scholar.google.com/scholar?q=Jakub+Piskorski%2C+Nicolas+Stefanovitch%2C+Giovanni+Da+San+Martino%2C+and+Preslav+Nakov%2E+2023%2E+Semeval%2D2023+task+3%3A+Detecting+the+category%2C+the+framing%2C+and+the+persuasion+techniques+in+online+news+in+a+mul)
19. Giancarlo Ruffo, Alfonso Semeraro, Anastasia Giachanou, and Paolo Rosso. 2023. Studying fake news spreading, polarisation dynamics, and manipulation by bots: A tale of networks and language. Computer science review, 47:100531. [↗](https://doi.org/10.1016/j.cosrev.2022.100531)
20. Aryan Sadeghi, Reza Alipour, Kamyar Taeb, Parimehr Morassafar, Nima Salemahim, and Ehsaneddin Asgari. 2023. Sinaai at semeval-2023 task 3: A multilingual transformer language model-based approach for the detection of news genre, framing and persuasion techniques. In Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 2168–2173. [find](https://scholar.google.com/scholar?q=Aryan+Sadeghi%2C+Reza+Alipour%2C+Kamyar+Taeb%2C+Parimehr+Morassafar%2C+Nima+Salemahim%2C+and+Ehsaneddin+Asgari%2E+2023%2E+Sinaai+at+semeval%2D2023+task+3%3A+A+multilingual+transformer+language+model%2Dbased+approach+for+)

*Updated 1 November 2026*