---
title: "AraBEM at WANLP 2022 Shared Task: Propaganda Detection in"
description: "This paper describes team AraBEM's entry to the WANLP 2022 shared task on detecting propaganda… 19 references, cited by 2 papers here."
authors: ["Eshrag Ali Refaee", "Basem Ahmed", "Motaz Saad"]
year: 2022
venue: "Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP)"
source: "https://aclanthology.org/2022.wanlp-1.62/"
doi: "10.18653/v1/2022.wanlp-1.62"
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/refaee-2022-arabem-wanlp-2022-shared-task/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/refaee-2022-arabem-wanlp-2022-shared-task/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2022](https://md.research.elamur.ai/free/en/papers/2022/page/7/) / AraBEM at WANLP 2022 Shared Task: Propaganda Detection in Arabic Tweets

# AraBEM at WANLP 2022 Shared Task: Propaganda Detection in Arabic Tweets

**Authors:** [Eshrag Ali Refaee](https://md.research.elamur.ai/free/en/authors/eshrag-refaee/), [Basem Ahmed](https://md.research.elamur.ai/free/en/authors/basem-ahmed/) and [Motaz Saad](https://md.research.elamur.ai/free/en/authors/motaz-saad/) · **Year:** 2022 · **Venue:** [Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP)](https://md.research.elamur.ai/free/en/venues/wanlp/) ([WANLP 2022](https://md.research.elamur.ai/free/en/venues/wanlp/2022/)) · **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/2022.wanlp-1.62/) · [PDF](https://aclanthology.org/2022.wanlp-1.62.pdf) · [DOI 10.18653/v1/2022.wanlp-1.62](https://doi.org/10.18653/v1/2022.wanlp-1.62) · [OpenAlex](https://openalex.org/W4385567184)

**Citations:** 2 citations (OpenAlex, 2026-10-14) · [2](https://md.research.elamur.ai/free/en/papers/refaee-2022-arabem-wanlp-2022-shared-task/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
This paper describes team AraBEM's entry to the WANLP 2022 shared task on detecting propaganda in Arabic tweets. The authors took part in subtask-1, where each tweet can carry several of 19 persuasion techniques, such as Loaded Language, Name-calling/Labeling, Straw Man or Whataboutism. The training set is tiny and badly skewed, and some techniques have no training examples at all.

The pipeline is simple:
- Clean the text with standard Arabic steps: letter normalisation, PyArabic tokenisation, removal of diacritics and elongations, and normalisation of URLs, user mentions and hashtags.
- Fine-tune a pre-trained BERT model with one independent yes/no output per technique.
- Mark a technique as present when its score exceeds 0.2.

The system reached a micro-F1 of 0.602 and placed third, against an organiser baseline of 0.079. The per-class breakdown shows that almost all of this score comes from the two or three most frequent techniques, and the macro average is only about 0.1. The work is therefore a useful reference point on how far a generic fine-tuned encoder gets on a small, imbalanced, fine-grained rhetorical-labelling task.

## Key points
- **Data:** 504 training tweets, 52 dev tweets and 323 test tweets, averaging about 15 tokens each. About a hundred training tweets carrying no technique were removed. Class imbalance was deliberately left untreated.
- **Label distribution:** Loaded Language (289 instances) and Name-calling/Labeling (186) dominate the training data, while several techniques have zero or a handful of examples.
- **Model setup:** a pre-trained BERT with a maximum sequence length of 70 tokens, chosen from tweet-length statistics. It was fine-tuned on train+dev, with a single decision threshold of 0.2 shared by all labels.
- **Overall results:** micro-F1 0.602 (third place), macro F1 about 0.1 and weighted F1 about 0.6.
- **Per-class results:** Loaded Language scores F1 0.83 and Name-calling 0.68, while most other classes score 0. The code is released on GitHub (motazsaad/Arabic-Proagenda-Detection).

## Interesting ideas
- Treating multi-label tagging as a set of independent binary decisions with a tuned (here low, 0.2) cut-off, and flagging per-label thresholds as the next step, is a cheap way to trade precision for recall on sparse labels.
- Reporting micro, macro and weighted averages next to per-class P/R/F1 shows that a respectable headline score can hide zero recall on most categories.
- Comparing against an organiser baseline (0.079) puts the system's number in context and works as a sanity check on task difficulty.
- Collapsing character elongations (e.g., "loooool") and interchangeable letter variants before tokenisation reduces sparsity in noisy social-media text.
- The task's taxonomy of persuasion techniques (Loaded Language, Straw Man, Whataboutism, Red Herring, Appeal to fear, and others) gives an explicit inventory of rhetorical moves rather than a single "propaganda" flag.

## For Mova

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

- **I1** (frames): detectors for speech acts, irony or implicature that may co-occur could each get their own calibrated threshold instead of one global argmax, and the chosen cut-off could be logged for explainability.
- **I2** (learn): evaluation of any multi-class component (frame roles, verb classes) should always print macro and per-class figures, so frequent classes cannot mask failures on rare ones.
- **I3** (learn): every reader or parser benchmark could ship with a trivial majority-label or random baseline as a negative control, so gains are measured against what imbalance alone yields.
- **I4** (tokens): the en tokenizer could map elongated or stylised spellings to a canonical lemma while keeping the original surface form for traceability.
- **I5** (frames): prag could add a layer of rhetorical-technique labels over speech acts. For example, Straw Man could be defined as misattributing a claim to another agent in the who-said-what frames.

*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

- [Jazan University](https://md.research.elamur.ai/free/en/universities/jazan-university/) — University, Gazan, Saudi Arabia · 9 works · 98 citations
- [Islamic University of Gaza](https://md.research.elamur.ai/free/en/universities/islamic-university-of-gaza/) — University, Gaza, Palestine · 3 works · 109 citations

Also on the byline: [Al-Aqsa University](https://ror.org/03e99kh24).

📚 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 · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,772 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: Eshrag Ali Refaee, Basem Ahmed and Motaz Saad. 2022. AraBEM at WANLP 2022 Shared Task: Propaganda Detection in Arabic Tweets. Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP).

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

```bibtex
@inproceedings{refaee2022arabem,
  title = {{AraBEM at WANLP 2022 Shared Task: Propaganda Detection in Arabic Tweets}},
  author = {Eshrag Ali Refaee, Basem Ahmed and Motaz Saad},
  booktitle = {Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP)},
  year = {2022},
  doi = {10.18653/v1/2022.wanlp-1.62},
  url = {https://doi.org/10.18653/v1/2022.wanlp-1.62}
}
```

- [← previous Ara-Women-Hate: An Annotated Corpus Dedicated to Hate Speech Detection against Women in the Arabic Community](https://md.research.elamur.ai/free/en/papers/guellil-2022-ara-women-hate-annotated-corpus/)
- [next → Are Current Task-oriented Dialogue Systems Able to Satisfy Impolite Users?](https://md.research.elamur.ai/free/en/papers/hu-2022-current-task-oriented-dialogue-systems/)
- [↑ back to papers of 2022](https://md.research.elamur.ai/free/en/papers/2022/page/7/)

## References

19 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Ehab A Abozinadah, Alex V Mbaziira, and J Jones. 2015. Detection of abusive accounts with arabic tweets. Int. J. Knowl. Eng.-IACSIT, 1(2):113–119. [↗](https://doi.org/10.7763/ijke.2015.v1.19)
2. Firoj Alam, Hamdy Mubarak, Wajdi Zaghouani, Preslav Nakov, and Giovanni Da San Martino. 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, Abu Dhabi, UAE. Association for Computational Linguistics.
3. Nuha Albadi, Maram Kurdi, and Shivakant Mishra. 2019. Investigating the effect of combining gru neural networks with handcrafted features for religious hatred detection on arabic twitter space. Social Network Analysis and Mining, 9(1):1–19. [↗](https://doi.org/10.1007/s13278-019-0587-5)
4. 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/). In LREC 2020 Workshop Language Resources and Evaluation Conference 11–16 May 2020, page 9. [↗](https://doi.org/10.48550/arxiv.2003.00104)
5. Adam Badawy and Emilio Ferrara. 2018. The rise of jihadist propaganda on social networks. Journal of Computational Social Science, 1(2):453–470. [↗](https://doi.org/10.1007/s42001-018-0015-z)
6. Akemi Takeoka Chatfield, Christopher G Reddick, and Uuf Brajawidagda. 2015. Tweeting propaganda, radicalization and recruitment: Islamic state supporters multi-sided twitter networks. In Proceedings of the 16th annual international conference on digital government research, pages 239–249. [find](https://scholar.google.com/scholar?q=Akemi+Takeoka+Chatfield%2C+Christopher+G+Reddick%2C+and+Uuf+Brajawidagda%2E+2015%2E+Tweeting+propaganda%2C+radicalization+and+recruitment%3A+Islamic+state+supporters+multi%2Dsided+twitter+networks%2E+In+Proceedings+o)
7. Deptii D Chaudhari and Ambika V Pawar. 2022. A systematic comparison of machine learning and nlp techniques to unveil propaganda in social media. Journal of Information Technology Research (JITR), 15(1):1– 14. [↗](https://doi.org/10.4018/jitr.299384)
8. Aditya Chetan, Brihi Joshi, Hridoy Sankar Dutta, and Tanmoy Chakraborty. 2019. Corerank: Ranking to detect users involved in blackmarket-based collusive retweeting activities. In Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining, pages 330–338. [find](https://scholar.google.com/scholar?q=Aditya+Chetan%2C+Brihi+Joshi%2C+Hridoy+Sankar+Dutta%2C+and+Tanmoy+Chakraborty%2E+2019%2E+Corerank%3A+Ranking+to+detect+users+involved+in+blackmarket%2Dbased+collusive+retweeting+activities%2E+In+Proceedings+of+the+Tw)
9. Mahdi Hashemi and Margeret Hall. 2019. Detecting and classifying online dark visual propaganda. Image and Vision Computing, 89:95–105. [↗](https://doi.org/10.1016/j.imavis.2019.06.001)
10. Mojtaba Heidarysafa, Kamran Kowsari, Tolu Odukoya, Philip Potter, Laura E Barnes, and Donald E Brown. 2020. Women in ISIS propaganda: a natural language processing analysis of topics and emotions in a comparison with a mainstream religious group. In Science and Information Conference, pages 610–624. Springer. [find](https://scholar.google.com/scholar?q=Mojtaba+Heidarysafa%2C+Kamran+Kowsari%2C+Tolu+Odukoya%2C+Philip+Potter%2C+Laura+E+Barnes%2C+and+Donald+E+Brown%2E+2020%2E+Women+in+ISIS+propaganda%3A+a+natural+language+processing+analysis+of+topics+and+emotions+in+a)
11. Giovanni Da San Martino, Stefano Cresci, Alberto Barrón-Cedeño, Seunghak Yu, Roberto Di Pietro, and Preslav Nakov. 2020. A survey on computational propaganda detection. arXiv preprint arXiv:2007.08024. [↗](https://arxiv.org/abs/2007.08024)
12. Subhadarshi Panda and Sarah Ita Levitan. 2021. [Detecting multilingual covid-19 misinformation on social media via contextualized embeddings](https://md.research.elamur.ai/free/en/papers/panda-2021-detecting-multilingual-covid-19-misinformation/). In Proceedings of the Fourth Workshop on NLP for Internet Freedom: Censorship, Disinformation, and Propaganda, pages 125–129.
13. Eshrag Refaee. 2017. Sentiment analysis for microblogging platforms in Arabic. In International conference on social computing and social media, pages 275–294. Springer, Cham. [find](https://scholar.google.com/scholar?q=Eshrag+Refaee%2E+2017%2E+Sentiment+analysis+for+microblogging+platforms+in+Arabic%2E+In+International+conference+on+social+computing+and+social+media%2C+pages+275%E2%80%93294%2E+Springer%2C+Cham%2E)
14. Eshrag A Refaee. 2021. A data-oriented approach for detecting offensive language in Arabic tweets. In 2021 International Conference on Software Engineering & Computer Systems and 4th International Conference on Computational Science and Information Management (ICSECS-ICOCSIM), pages 244–248. IEEE. [find](https://scholar.google.com/scholar?q=Eshrag+A+Refaee%2E+2021%2E+A+data%2Doriented+approach+for+detecting+offensive+language+in+Arabic+tweets%2E+In+2021+International+Conference+on+Software+Engineering+%26+Computer+Systems+and+4th+International+Con)
15. George-Alexandru Vlad, Mircea-Adrian Tanase, Cristian Onose, and Dumitru-Clementin Cercel. 2019. Sentence-level propaganda detection in news articles with transfer learning and bert-bilstm-capsule model. In Proceedings of the second workshop on natural language processing for internet freedom: Censorship, Disinformation, and Propaganda, pages 148–154. [find](https://scholar.google.com/scholar?q=George%2DAlexandru+Vlad%2C+Mircea%2DAdrian+Tanase%2C+Cristian+Onose%2C+and+Dumitru%2DClementin+Cercel%2E+2019%2E+Sentence%2Dlevel+propaganda+detection+in+news+articles+with+transfer+learning+and+bert%2Dbilstm%2Dcapsule+mod)
16. Taha Zerrouki. 2010. PyArabic, an Arabic language library for python. [find](https://scholar.google.com/scholar?q=Taha+Zerrouki%2E+2010%2E+PyArabic%2C+an+Arabic+language+library+for+python%2E)
17. Emilio Ferrara. 2017. Contagion dynamics of extremist propaganda in social networks. Information Sciences, 418:1–12. [↗](https://doi.org/10.1016/j.ins.2017.07.030)
18. Ivan Habernal, Raffael Hannemann, Christian Pollak, Christopher Klamm, Patrick Pauli, and Iryna Gurevych. 2017. Argotario: Computational argumentation meets serious games. arXiv preprint arXiv:1707.06002. [↗](https://arxiv.org/abs/1707.06002)
19. Fouzi Harrag, Maria Debbah, Kareem Darwish, and Ahmed Abdelali. 2021. [Bert transformer model for detecting arabic gpt2 auto-generated tweets](https://md.research.elamur.ai/free/en/papers/harrag-2020-bert-transformer-model-detecting-arabic/). arXiv preprint arXiv:2101.09345. [↗](https://arxiv.org/abs/2101.09345)

*Updated 1 November 2026*