Elamur Research⌕
☰
home / papers / 2026 / The NakbaVirality Shared Task on Multimodal Virality Prediction in High-Stakes Discourse

The NakbaVirality Shared Task on Multimodal Virality Prediction in High-Stakes Discourse

Paper · 2026 · Saad Ezzini, Salima Lamsiyah, Shadi Abudalfa et al.

Summary

NakbaVirality is a shared task, run at the Nakba-NLP workshop (LREC 2026), on predicting whether a social media post about the post-October-7-2023 conflict will become viral. The prediction must come from the content alone: the post's text and its image. Earlier virality models mostly relied on early likes, reposts or diffusion graphs, and those signals do not exist when a post is first published. The organizers built a corpus of 2,600 anonymized X and Reddit posts. Each post was found by multilingual keyword search, had to include at least one image, and was deduplicated by platform identifiers. Engagement counts were normalized within each platform and then binned into low, medium and high virality. The organizers supplied standardized splits, transformer and convolutional baselines, and scoring by macro-F1 and accuracy. Eighteen teams registered and four made official test submissions. The winning system did not use deep multimodal fusion. It used a gradient-boosted classifier built on a smoothed encoding of the source community plus shallow text features. This suggests that where a post appears matters more for its reach than what it says. The authors present the task as the first benchmark of its kind for Arabic high-stakes discourse.

Key points

  • The dataset has 2,600 posts from X and Reddit, all published from 2023-10-07 (UTC) onward. Each post comes with text, an image and engagement-derived labels in three classes: low, medium and high.
  • About 52.8% of the posts are in English and 8.9% in Arabic. The remainder includes Farsi, Hebrew, Spanish and Turkish, so the "Arabic" framing describes the topic better than the language mix.
  • Engagement is computed as E = likes + reposts + replies. Labels come from normalizing E per platform rather than comparing raw counts across X and Reddit. Raw engagement fields are kept so the labels can be reproduced.
  • The winner, HCMUS_TheFangs, combined three kinds of input in an XGBoost classifier:
    • a Bayesian-smoothed target encoding of the source community;
    • character, word and hashtag counts;
    • TF-IDF text features. This system outperformed all deep learning entries.
  • The authors list several limitations. User identities and follower graphs were removed for privacy, which rules out cascade analysis. Truly viral posts are rare, so the classes are heavily imbalanced.

Interesting ideas

  • I1. Normalizing scores inside each source before binning them into shared classes prevents one source's larger raw scale from dominating the labels.
  • I2. In this task a cheap metadata-plus-counts model beat deep multimodal systems, which shows that a dataset can contain a strong shortcut feature unrelated to content.
  • I3. Bayesian-smoothed target encoding replaces a sparse category with its outcome rate, pulled toward the global prior when the category has few examples. This keeps rare categories from receiving extreme values.
  • I4. The task keeps the raw platform fields alongside the derived labels, so the label rule can be changed and the labels rebuilt without re-crawling.
  • I5. The task reports macro-F1 alongside accuracy because the class of interest (viral posts) is a small minority.

For Mova

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

  • I1 (learn): Witness parsers' confidence scores could be rescaled per parser before deciding silver-data agreement, so that one overconfident parser does not outvote the others.
  • I2 (learn): Every evaluation could include a shallow "metadata only" negative-control baseline; if it scores close to the full pipeline, the benchmark is leaking a shortcut.
  • I3 (tokens): Rare suffixes, unknown-word shapes or rare lemmas in the tagger could get smoothed tag-distribution features instead of noisy raw counts.
  • I4 (learn): Storing every witness parser's raw output next to the agreed silver tree would let the agreement rule be revised and the silver data regenerated cheaply.
  • I5 (answer): School-test and QA results could be reported as macro scores over question types, so that rare types such as negation or counting questions are not hidden by the common ones.

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

🎓King Fahd University of Petroleum and MineralsDhahran · Saudi Arabia52 works · 4,666 citations🎓University of LuxembourgLuxembourg · Luxembourg27 works · 337 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksSocial Sciences · 1,772 worksSociology and Political Science · 577 works

🧭 Topics 2Research subjects and languages the work is about.Vision & language · 657 worksDiscourse analysis · 466 works

Cite

Cite as Saad Ezzini, Salima Lamsiyah, Shadi Abudalfa, Samir El-Amrany and Walid Alsafadi. 2026. The NakbaVirality Shared Task on Multimodal Virality Prediction in High-Stakes Discourse. Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/3qkodh8922a6

@inproceedings{ezzini2026nakbavirality,
  title = {{The NakbaVirality Shared Task on Multimodal Virality Prediction in High-Stakes Discourse}},
  author = {Saad Ezzini, Salima Lamsiyah, Shadi Abudalfa, Samir El-Amrany and Walid Alsafadi},
  booktitle = {Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026},
  year = {2026},
  doi = {10.63317/3qkodh8922a6},
  url = {https://doi.org/10.63317/3qkodh8922a6}
}

References

25 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jonah Berger and Katherine L Milkman. 2012. What makes online content viral? Journal of marketing research, 49(2):192–205. ↗
  2. Adam Bielski and Tomasz Trzcinski. 2018. Pay attention to virality: understanding popularity of social media videos with the attention mechanism. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pages 2335–2337. find
  3. Ming Cheung, James She, Alvin Junus, and Lei Cao. 2016. Prediction of virality timing using cascades in social media. ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 13(1):1–23. ↗
  4. Jacob Devlin. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805. ↗
  5. Benjamin Doerr, Mahmoud Fouz, and Tobias Friedrich. 2012. Why rumors spread so quickly in social networks. Communications of the ACM, 55(6):70–75. ↗
  6. Samir El-amrany, Matthias R. Brust, Salima Lamsiyah, and Pascal Bouvry. 2025. Reddit-V: A virality prediction dataset and zero-shot evaluation with large language models. In Proceedings of the 15th International Conference on Recent Advances in Natural Language Processing - Natural Language Processing in the Generative AI Era, pages 334–341, Varna, Bulgaria. INCOMA Ltd., Shoumen, Bulgaria. find
  7. Mercedes Esteban-Bravo, Jose M Vidal-Sanz, et al. 2024. Predicting the virality of fake news at the early stage of dissemination. Expert Systems with Applications, 248:123390. ↗
  8. Liqun Gao, Yujia Liu, Hongwu Zhuang, Haiyang Wang, Bin Zhou, and Aiping Li. 2021. Public opinion early warning agent model: A deep learning cascade virality prediction model based on multi-feature fusion. Frontiers in Neurorobotics, 15:674322. ↗
  9. Marco Guerini, Carlo Strapparava, and Gozde Ozbal. 2011. Exploring text virality in social networks. In proceedings of the international AAAI conference on web and social media, volume 5, pages 506–509. find
  10. Tuan-Anh Hoang and Ee-Peng Lim. 2016. Tracking virality and susceptibility in social media. In Proceedings of the 25th ACM international on conference on information and knowledge management, pages 1059–1068. find
  11. Mustafa Jarrar, Mo El-Haj, Amal Haddad, Serin Atiani, Shadi Abudalfa, Khalil Sima’an, Paul Rayson, and Camille Mansour, editors. 2026. Proceedings of the second International Workshop on Nakba Narratives as Language Resources. Association for Computational Linguistics, Spain. find
  12. Damian Konrad Kowalczyk and Jan Larsen. 2018. Scalable privacy-compliant virality prediction on twitter. arXiv preprint arXiv:1812.06034. ↗
  13. Xiaoyan Lu and Boleslaw K Szymanski. 2018. Scalable prediction of global online media news virality. IEEE Transactions on Computational Social Systems, 5(3):858–870. ↗
  14. Christian E Maldonado-Sifuentes, Jason Angel, Grigori Sidorov, Olga Kolesnikova, and Alexander Gelbukh. 2021. Virality prediction for news tweets using roberta. In Mexican International Conference on Artificial Intelligence, pages 81– 95. Springer. find
  15. Maryam Mousavi, Hasan Davulcu, Mohsen Ahmadi, Robert Axelrod, Richard Davis, and Scott Atran. 2022. Effective messaging on social media: What makes online content go viral? In Proceedings of the ACM web conference 2022, pages 2957–2966. find
  16. David Powers. 2011. Evaluation: From precision, recall and f-measure to roc, informedness, markedness & correlation. Journal of Machine Learning Technologies, 2(1):37–63. find
  17. Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. Learning transferable visual models from natural language supervision. arXiv preprint arXiv:2103.00020. ↗
  18. Rikaz Rameez, Hossein A Rahmani, and Emine Yilmaz. 2022. Viralbert: A user focused bert-based approach to virality prediction. In Adjunct Proceedings of the 30th ACM Conference on User Modeling, Adaptation and Personalization, pages 85–89. find
  19. Christoffer Riis, Damian Konrad Kowalczyk, and Lars Kai Hansen. 2020. On the limits to multimodal popularity prediction on instagram–a new robust, efficient and explainable baseline. arXiv preprint arXiv:2004.12482. ↗
  20. David Shamma, Jude Yew, Lyndon Kennedy, and Elizabeth Churchill. 2011. Viral actions: Predicting video view counts using synchronous sharing behaviors. In Proceedings of the International AAAI Conference on Web and Social Media, volume 5, pages 618–621. find
  21. Shivangi Singhal, Rajiv Ratn Shah, Tanmoy Chakraborty, Ponnurangam Kumaraguru, and Shin’ichi Satoh. 2019. Spotfake: A multi-modal framework for fake news detection. In 2019 IEEE fifth international conference on multimedia big data (BigMM), pages 39–47. IEEE. find
  22. Sho Tsugawa and Hiroyuki Ohsaki. 2017. On the relation between message sentiment and its virality on social media. Social network analysis and mining, 7(1):19. ↗
  23. Lilian Weng, Filippo Menczer, and Yong-Yeol Ahn. 2013. Virality prediction and community structure in social networks. Scientific reports, 3(1):1–6. ↗
  24. Zhixuan Xu and Minghui Qian. 2023. Predicting popularity of viral content in social media through a temporal-spatial cascade convolutional learning framework. Mathematics, 11(14):3059. ↗
  25. Xuan Zhang and Wei Gao. 2024. Predicting viral rumors and vulnerable users with graph-based neural multi-task learning for infodemic surveillance. Information Processing & Management, 61(1):103520. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ezzini-2026-nakbavirality-shared-task-multimodal-virality/