---
title: "MemeMind at ArAIEval Shared Task — paper by Shah et al. (2024)"
description: "The paper describes the MemeMind team's system for ArAIEval 2024 Task 2. 24 references, cited by 9 papers here."
authors: ["Uzair Shah", "Md. Rafiul Biswas", "Marco Agus", "Mowafa Househ", "Wajdi Zaghouani"]
year: 2024
venue: "Proceedings of the Second Arabic Natural Language Processing Conference"
source: "https://aclanthology.org/2024.arabicnlp-1.45/"
doi: "10.18653/v1/2024.arabicnlp-1.45"
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/shah-2024-mememind-araieval-shared-task-generative/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/shah-2024-mememind-araieval-shared-task-generative/)

[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/9/) / MemeMind at ArAIEval Shared Task: Generative Augmentation and Feature Fusion for Multimodal Propaganda Detection in Arabic Memes through Advanced Language and Vision Models

# MemeMind at ArAIEval Shared Task: Generative Augmentation and Feature Fusion for Multimodal Propaganda Detection in Arabic Memes through Advanced Language and Vision Models

**Authors:** [Uzair Shah](https://md.research.elamur.ai/free/en/authors/uzair-shah/), [Md. Rafiul Biswas](https://md.research.elamur.ai/free/en/authors/md-rafiul-biswas/), Marco Agus, Mowafa Househ and [Wajdi Zaghouani](https://md.research.elamur.ai/free/en/authors/wajdi-zaghouani/) · **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-1.45/) · [PDF](https://aclanthology.org/2024.arabicnlp-1.45.pdf) · [DOI 10.18653/v1/2024.arabicnlp-1.45](https://doi.org/10.18653/v1/2024.arabicnlp-1.45) · [OpenAlex](https://openalex.org/W4402671487)

**Citations:** 3 citations (OpenAlex, 2026-10-14) · [9](https://md.research.elamur.ai/free/en/papers/shah-2024-mememind-araieval-shared-task-generative/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper describes the MemeMind team's system for ArAIEval 2024 Task 2. The task asks whether Arabic memes are propagandistic, judged three ways: from the meme text alone (2A), from the image alone (2B), and from both together (2C). The training data is skewed toward non-propaganda, and the team's main response is to create extra examples with generative models.

- **Text (2A):** several Arabic encoders were fine-tuned with class weighting, using additional meme-like texts written by ChatGPT4.
- **Images (2B):** ResNet18, EfficientFormerV2 and ConvNeXt-tiny were trained, adding new propaganda images that DALL-E-2 produced from real memes in its image-to-image mode.
- **Multimodal (2C):** frozen AraBERT and ConvNeXt-tiny encoders supplied embeddings, which were concatenated and passed to a trainable fusion layer and a binary classifier.

On the development set, augmentation helped both single-modality models, and fusion scored best overall at macro F1 0.837. On the official leaderboard the team placed near the top in all three subtasks. The work matters because propaganda detection in Arabic memes has received little attention compared with English.

## Key points
- **Data (Alam et al., 2024):** 2143 / 312 / 607 text–image pairs for train / validation / test. The training split has 603 propaganda and 1540 non-propaganda items.
- **Text results (dev macro F1, AraBERT):**
  - 0.763 with no augmentation;
  - 0.789 with synthetic propaganda texts only;
  - 0.80 when both classes were augmented, doubling the data.
- **Image and fusion results (dev macro F1):**
  - ConvNeXt-tiny rose only from 0.729 to 0.736 with DALL-E-2 images;
  - a Random Forest over concatenated features reached 0.819;
  - the learned fusion layer reached 0.837.
- **Official test scores:**
  - 2A: 0.7464 (third);
  - 2B: 0.6634 (second);
  - 2C: 0.7972 (fourth, in a tight cluster just behind the top three teams).
- **Training setup:** AdamW, learning rate 1e-5 for text and fusion and 1e-4 for vision, batch size 32, 128 tokens, 224x224 images. Dropout was raised step by step from 0.25 to 0.5 to 0.75 as the share of synthetic data grew.

## Interesting ideas
- Filling the minority class with LLM-written examples, then doubling both classes, gave stepwise gains, so synthetic data can repair skew rather than only add volume.
- The authors raised regularization (dropout, drop-path) as the synthetic portion grew, which guards against memorizing the generator's style.
- A new classification head is trained first on a frozen backbone, and the full network is unfrozen only after that, so random head weights cannot damage pretrained features.
- Before training a neural fusion layer, the team tested a simple Random Forest over concatenated frozen embeddings, and it already came close to the fusion score.
- The image model's small dev gain from synthetic images was followed by a much lower test score, a warning that generated data can make development results look better than they are.

## For Mova

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

- **I1** (learn): rare UD constructions such as clefts or gapping could get LLM-written sentences that enter silver data only after the witness parsers agree on their trees.
- **I2** (learn): when gparse training mixes in more LLM-teacher silver, dropout could scale with the silver-to-gold ratio and be checked against gold-only dev LAS.
- **I3** (tree): when gparse gets a new label head for a converted UD dialect, warming up the head on frozen encoder layers could protect existing LAS.
- **I4** (frames): irony or speech-act detection could first combine tree features and VerbNet role features in an inspectable tree-based model, keeping the decision explainable before anything neural is tried.
- **I5** (learn): any augmentation should be judged only on held-out real text, with a negative control trained on shuffled or synthetic-only data.

*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

- [Hamad bin Khalifa University](https://md.research.elamur.ai/free/en/universities/hamad-bin-khalifa-university/) — University, Doha, Qatar · 141 works · 14,175 citations
- [Qatar Foundation](https://md.research.elamur.ai/free/en/ngos/qatar-foundation/) — Nonprofit, Doha, Qatar · 17 works · 2,111 citations
- [Northwestern University in Qatar](https://md.research.elamur.ai/free/en/universities/northwestern-university-in-qatar/) — University, Doha, Qatar · 24 works · 148 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 · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 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 (2): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works · [🖼️ Vision & language](https://md.research.elamur.ai/free/en/topics/multimodal/) · 657 works

## Cite

Cite as: Uzair Shah, Md. Rafiul Biswas, Marco Agus, Mowafa Househ and Wajdi Zaghouani. 2024. MemeMind at ArAIEval Shared Task: Generative Augmentation and Feature Fusion for Multimodal Propaganda Detection in Arabic Memes through Advanced Language and Vision Models. 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{shah2024mememind,
  title = {{MemeMind at ArAIEval Shared Task: Generative Augmentation and Feature Fusion for Multimodal Propaganda Detection in Arabic Memes through Advanced Language and Vision Models}},
  author = {Uzair Shah, Md. Rafiul Biswas, Marco Agus, Mowafa Househ and Wajdi Zaghouani},
  booktitle = {Proceedings of the Second Arabic Natural Language Processing Conference},
  year = {2024},
  doi = {10.18653/v1/2024.arabicnlp-1.45},
  url = {https://doi.org/10.18653/v1/2024.arabicnlp-1.45}
}
```

- [← previous Maximize Your Data's Potential: Enhancing LLM Accuracy with Two-Phase Pretraining](https://md.research.elamur.ai/free/en/papers/feng-2024-maximize-data-s-potential-enhancing/)
- [next → MiniPLM: Knowledge Distillation for Pre-Training Language Models](https://md.research.elamur.ai/free/en/papers/gu-2024-miniplm-knowledge-distillation-pre-training/)
- [↑ back to papers of 2024](https://md.research.elamur.ai/free/en/papers/2024/page/9/)

## References

24 references, 6 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. 2021. ArBERT: Bert for arabic language understanding. arXiv preprint arXiv:2110.05609. [↗](https://arxiv.org/abs/2110.05609)
2. Samer Al-Saqqa and Arafat Awajan. 2021. Memeganda: Identifying propaganda memes using encoding and decoding deep features. IEEE Access, 9:153670– 153682. [find](https://scholar.google.com/scholar?q=Samer+Al%2DSaqqa+and+Arafat+Awajan%2E+2021%2E+Memeganda%3A+Identifying+propaganda+memes+using+encoding+and+decoding+deep+features%2E+IEEE+Access%2C+9%3A153670%E2%80%93+153682%2E)
3. Firoj Alam, Abul Hasnat, Fatema Ahmed, Md Arid Hasan, and Maram Hasanain. 2024. [ArMeme: Propagandistic content in arabic memes](https://md.research.elamur.ai/free/en/papers/alam-2024-armeme-propagandistic-content-arabic-memes/). arXiv: 2406.03916. [↗](https://arxiv.org/abs/2406.03916)
4. Maram Hasanain, Firoj Alam, Hamdy Mubarak, Samir Abdaljalil, Wajdi Zaghouani, Preslav Nakov, Giovanni Da San Martino, and Abed Alhakim Freihat. 2023b. [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. Maram Hasanain, Md. Arid Hasan, Fatema Ahmed, Reem Suwaileh, Md.
5. Rafiul Biswas, Wajdi Zaghouani, and Firoj Alam. 2024b. [ArAIEval Shared Task: Propagandistic techniques detection in unimodal and multimodal arabic content](https://md.research.elamur.ai/free/en/papers/hasanain-2024-araieval-shared-task-propagandistic-techniques/). In Proceedings of the Second Arabic Natural Language Processing Conference (ArabicNLP 2024), Bangkok. Association for Computational Linguistics.
6. 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.
7. Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2015. [Deep residual learning for image recognition](https://md.research.elamur.ai/free/en/papers/he-2015-deep-residual-learning-image-recognition/). arXiv preprint arXiv:1512.03385. [↗](https://arxiv.org/abs/1512.03385)
8. Tariq Alhindi, Savvas Petridis, and Smaranda Muresan. 2019. Arpr: Arabic propaganda detection using transfer learning. In Proceedings of the 2nd Workshop on NLP for Internet Freedom: Censorship, Disinformation, and Propaganda (NLP4IF), pages 1–6. [find](https://scholar.google.com/scholar?q=Tariq+Alhindi%2C+Savvas+Petridis%2C+and+Smaranda+Muresan%2E+2019%2E+Arpr%3A+Arabic+propaganda+detection+using+transfer+learning%2E+In+Proceedings+of+the+2nd+Workshop+on+NLP+for+Internet+Freedom%3A+Censorship%2C+Disin)
9. Garth S Jowett and Victoria O’donnell. 2018. Propaganda & persuasion. Sage Publications. [find](https://scholar.google.com/scholar?q=Garth+S+Jowett+and+Victoria+O%E2%80%99donnell%2E+2018%2E+Propaganda+%26+persuasion%2E+Sage+Publications%2E)
10. Belal Alkhamissi, Mohamed Gabr, Marwan Khaled, and Rania Essam. 2022. Adapting multilingual neural language models for arabic dialects. arXiv preprint arXiv:2202.06741. [↗](https://arxiv.org/abs/2202.06741)
11. 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)
12. Alexander Chernyavskiy, Dmitry Ilvovsky, and Preslav Nakov. 2024. Unleashing the power of discourse-enhanced transformers for propaganda detection. In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1452–1462. [find](https://scholar.google.com/scholar?q=Alexander+Chernyavskiy%2C+Dmitry+Ilvovsky%2C+and+Preslav+Nakov%2E+2024%2E+Unleashing+the+power+of+discourse%2Denhanced+transformers+for+propaganda+detection%2E+In+Proceedings+of+the+18th+Conference+of+the+Europea)
13. Naoya Inoue, Pontus Stenetorp, and Kentaro Inui. 2021. Interplay between preferences and machine learning in language model fine-tuning. arXiv preprint arXiv:2110.08413. [↗](https://arxiv.org/abs/2110.08413)
14. Douwe Kiela, Hamed Firooz, Aravind Mohan, Vedanuj Goswami, Amanpreet Singh, Pratik Ringshia, and Davide Testuggine. 2020. Hateful memes challenge: Detecting hate speech in multimodal memes. In Advances in Neural Information Processing Systems, volume 33, pages 2611–2624. [find](https://scholar.google.com/scholar?q=Douwe+Kiela%2C+Hamed+Firooz%2C+Aravind+Mohan%2C+Vedanuj+Goswami%2C+Amanpreet+Singh%2C+Pratik+Ringshia%2C+and+Davide+Testuggine%2E+2020%2E+Hateful+memes+challenge%3A+Detecting+hate+speech+in+multimodal+memes%2E+In+Advance)
15. Vasiliki Kougia, Simon Fetzel, Thomas Kirchmair, Erion Çano, Sina Moayed Baharlou, Sahand Sharifzadeh, and Benjamin Roth. 2023. Memegraphs: Linking memes to knowledge graphs. In International Conference on Document Analysis and Recognition, pages 534–551. Springer. [find](https://scholar.google.com/scholar?q=Vasiliki+Kougia%2C+Simon+Fetzel%2C+Thomas+Kirchmair%2C+Erion+%C3%87ano%2C+Sina+Moayed+Baharlou%2C+Sahand+Sharifzadeh%2C+and+Benjamin+Roth%2E+2023%2E+Memegraphs%3A+Linking+memes+to+knowledge+graphs%2E+In+International+Conferen)
16. Yanyu Li, Ju Hu, Yang Wen, Georgios Evangelidis, Kamyar Salahi, Yanzhi Wang, Sergey Tulyakov, and Jian Ren. 2023. Rethinking vision transformers for mobilenet size and speed. In Proceedings of the IEEE international conference on computer vision. [find](https://scholar.google.com/scholar?q=Yanyu+Li%2C+Ju+Hu%2C+Yang+Wen%2C+Georgios+Evangelidis%2C+Kamyar+Salahi%2C+Yanzhi+Wang%2C+Sergey+Tulyakov%2C+and+Jian+Ren%2E+2023%2E+Rethinking+vision+transformers+for+mobilenet+size+and+speed%2E+In+Proceedings+of+the+IEE)
17. Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, and Saining Xie. 2022. A convnet for the 2020s. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). [↗](https://doi.org/10.48550/arxiv.2201.03545)
18. Muhammad Arslan Manzoor, Sarah Albarri, Ziting Xian, Zaiqiao Meng, Preslav Nakov, and Shangsong Liang. 2023. Multimodality representation learning: A survey on evolution, pretraining and its applications. ACM Transactions on Multimedia Computing, Communications and Applications, 20(3):1–34. [↗](https://doi.org/10.1145/3617833)
19. Francisco Rangel, Paolo Rosso, Anis Charfi, and Wajdi Zaghouani. 2019. Detecting deceptive tweets in arabic for cyber-security. In 2019 IEEE International Conference on Intelligence and Security Informatics (ISI), pages 86–91. IEEE. [find](https://scholar.google.com/scholar?q=Francisco+Rangel%2C+Paolo+Rosso%2C+Anis+Charfi%2C+and+Wajdi+Zaghouani%2E+2019%2E+Detecting+deceptive+tweets+in+arabic+for+cyber%2Dsecurity%2E+In+2019+IEEE+International+Conference+on+Intelligence+and+Security+Infor)
20. Paolo Rosso, Francisco Rangel, Iraisy Hernández Far’ıas, Leticia Cagnina, Wajdi Zaghouani, and Anis Charfi. 2018. A survey on author profiling, deception, and irony detection for the arabic language. Language and Linguistics Compass, 12(4):e12275. [↗](https://doi.org/10.1111/lnc3.12275)
21. Uzair Shah, Muhammad Tukur, Mahmood Alzubaidi, Giovanni Pintore, Enrico Gobbetti, Mowafa Househ, Jens Schneider, and Marco Agus. 2024. Multi-panowise: Holistic deep architecture for multi-task dense prediction from a single panoramic image. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 1311–1321. [find](https://scholar.google.com/scholar?q=Uzair+Shah%2C+Muhammad+Tukur%2C+Mahmood+Alzubaidi%2C+Giovanni+Pintore%2C+Enrico+Gobbetti%2C+Mowafa+Househ%2C+Jens+Schneider%2C+and+Marco+Agus%2E+2024%2E+Multi%2Dpanowise%3A+Holistic+deep+architecture+for+multi%2Dtask+dense+p)
22. Anna Wilson, Seb Wilkes, Yayoi Teramoto, and Scott Hale. 2023. Multimodal analysis of disinformation and misinformation. Royal Society Open Science, 10(12):230964. [↗](https://doi.org/10.1098/rsos.230964)
23. Wajdi Zaghouani. 2018. A large-scale social media corpus for the detection of youth depression (project note). Procedia Computer Science, 142:347–351. [find](https://scholar.google.com/scholar?q=Wajdi+Zaghouani%2E+2018%2E+A+large%2Dscale+social+media+corpus+for+the+detection+of+youth+depression+%28project+note%29%2E+Procedia+Computer+Science%2C+142%3A347%E2%80%93351%2E)
24. Wajdi Zaghouani and Anis Charfi. 2018. Arap-tweet: A large multi-dialect twitter corpus for gender, age and language variety identification. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018). [find](https://scholar.google.com/scholar?q=Wajdi+Zaghouani+and+Anis+Charfi%2E+2018%2E+Arap%2Dtweet%3A+A+large+multi%2Ddialect+twitter+corpus+for+gender%2C+age+and+language+variety+identification%2E+In+Proceedings+of+the+Eleventh+International+Conference+on+)

*Updated 1 November 2026*