Elamur Research⌕
☰
home / papers / 2026 / The Argonauts at SemEval-2026 Task 9: Multilingual Polarization Detection and Classification Using LLM Prompting and Transformer Fine-Tuning

The Argonauts at SemEval-2026 Task 9: Multilingual Polarization Detection and Classification Using LLM Prompting and Transformer Fine-Tuning

Paper · 2026 · Sha Newaz Mahmud, Sajib Bhattacharjee, Md. Refaj Hossan et al.

Summary

The authors take part in SemEval-2026 Task 9, which concerns automatic recognition of polarizing social media content. Subtask 1 is a yes/no decision on whether a text is polarized. Subtask 2 tags polarized texts with up to five types: political/ideological, racial/ethnic, religious, gender/sexual and other. The team uses the English and Bengali portions of the POLAR benchmark.

For the binary task, they prompt a 4-bit quantized Qwen3-14B with a structured few-shot template. This outperforms both a fine-tuned XLM-RoBERTa-base and the RemBERT baseline supplied by the organizers.

For type classification, they fine-tune encoders such as XLM-RoBERTa-large and RoBERTa-base. The setup combines asymmetric loss, MarianMT back-translation to enlarge rare classes, and decision thresholds tuned separately for each label.

The prompting approach places near the top for Bengali. The multi-label systems beat the baseline but lose much of their development-set performance on the test set. The paper's practical lessons are that LLM prompting suits coarse polarization judgments, while per-label threshold tuning is fragile when label distributions shift.

Key points

  • Data: POLAR covers 22 languages; only English and Bengali were used.
    • English: 3222 train / 160 dev / 1452 test.
    • Bengali: 3333 train / 166 dev / 1501 test.
    • In the English training data, the gender/sexual label is extremely rare (about 43.8:1).
  • Subtask 1 (test macro-F1): Qwen3-14B few-shot prompting reached 0.847 for Bengali (4th place) and 0.808 for English (9th).
    • XLM-RoBERTa-base: 0.821 / 0.762.
    • RemBERT baseline: 0.754 / 0.694.
  • Subtask 2 setup:
    • Asymmetric loss with γ+ = 1 and γ− = 4.
    • Back-translation via French, German or Hindi, keeping only outputs with at least 50% token overlap with the original.
    • Per-label threshold search, which added up to 24.6 dev macro-F1 points.
  • Subtask 2 results (test macro-F1): 0.454 for English (21st) and 0.206 for Bengali (26th), versus RemBERT at 0.385 and 0.194.
  • Dev-to-test collapse:
    • XLM-RoBERTa-large fell from 0.827 to 0.454; BanglaBERT fell from 0.921 to 0.189.
    • The authors attribute this to thresholds overfitted to the dev set and a genuine shift in label distribution between dev and test.
    • Few-shot Qwen3 was also tried on Subtask 2 and dropped from 0.769 to 0.381.

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

🎓Chittagong University of Engineering & TechnologyChittagong · Bangladesh25 works · 29 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Multilingual & cross-lingual NLP · 828 works

Cite

Cite as Sha Newaz Mahmud, Sajib Bhattacharjee, Md. Refaj Hossan, Kawsar Ahmed and Mohammed Moshiul Hoque. 2026. The Argonauts at SemEval-2026 Task 9: Multilingual Polarization Detection and Classification Using LLM Prompting and Transformer Fine-Tuning. Proceedings of the 20th International Workshop on Semantic Evaluation (2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.semeval-1.366

@inproceedings{mahmud2026argonauts,
  title = {{The Argonauts at SemEval-2026 Task 9: Multilingual Polarization Detection and Classification Using LLM Prompting and Transformer Fine-Tuning}},
  author = {Sha Newaz Mahmud, Sajib Bhattacharjee, Md. Refaj Hossan, Kawsar Ahmed and Mohammed Moshiul Hoque},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.366},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.366}
}

References

17 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Christopher A. Bail, Lisa P. Argyle, Taylor W. Brown, John P. Bumpus, Haohan Chen, M. B. Fallin Hunzaker, Jaemin Lee, Marcus Mann, Friedolin Merhout, and Alexander Volfovsky. 2018. Exposure to opposing views on social media can increase political polarization. Proceedings of the National Academy of… find
  2. Valerio Basile, Cristina Bosco, Elisabetta Fersini, Debora Nozza, Viviana Patti, Francisco Manuel Rangel Pardo, Paolo Rosso, and Manuela Sanguinetti. 2019. SemEval-2019 Task 5: Multilingual detection of hate speech against immigrants and women in Twitter. In Proceedings of SemEval-2019, pages 54– 63…
  3. Abhik Bhattacharjee, Tahmid Hasan, Wasi Uddin Ahmad, Kazi Samin Mubasshir, Md Saiful Islam, Anindya Iqbal, M. Sohel Rahman, and Rifat Shahriyar. 2022. BanglaBERT: Language model pretraining and benchmarks for low-resource language understanding evaluation in Bangla. In Findings of NAACL, pages 1318–…
  4. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. Proceedings of ACL, pages 8440–8451. find
  5. Michael D. Conover, Jacob Ratkiewicz, Matthew Francisco, Bruno Gonçalves, Filippo Menczer, and Alessandro Flammini. 2011. Political polarization on Twitter. In Proceedings of the Fifth International AAAI Conference on Weblogs and Social Media. find
  6. Paul DiMaggio, John Evans, and Bethany Bryson. 1996. Have Americans’ social attitudes become more polarized? American Journal of Sociology, 102(3):690– 755. find
  7. Kiran Garimella, Gianmarco De Francisci Morales, Aristides Gionis, and Michael Mathioudakis. 2018. Quantifying controversy on social media. ACM Transactions on Social Computing, 1(1):1–27. find
  8. Debanjan Ghosh, Kai North, Ekaterina Kochmar, Mamoru Komachi, and Marcos Zampieri, editors. 2026. Proceedings of the 20th International Workshop on Semantic Evaluation (SemEval-2026). Association for Computational Linguistics, San Diego, United States. find
  9. Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. 2021. DeBERTa: Decoding-enhanced BERT with disentangled attention. In Proceedings of ICLR.
  10. Ruibo Jiang, Xinning Hua, Valerie Berger, Samuel Barkley, Jonathan Bernstein, Lu Xiao, and Jiebo Luo. 2022. CommunityLM: Probing partisan worldviews from language models. arXiv preprint arXiv:2209.07065. find
  11. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. RoBERTa: A robustly optimized BERT pretraining approach. arXiv preprint arXiv:1907.11692.
  12. Saif Mohammad, Svetlana Kiritchenko, Parinaz Sobhani, Xiaodan Zhu, and Colin Cherry. 2016. SemEval-2016 Task 6: Detecting stance in tweets. In Proceedings of SemEval-2016, pages 31–41.
  13. Usman Naseem, Robert Geislinger, Juan Ren, Sarah Kohail, Rudy Garrido Veliz, P Sam Sahil, Yiran Zhang, Marco Antonio Stranisci, Idris Abdulmumin, Özge Alacam, Cengiz Acarürk, Aisha Jabr, Saba Anwar, Abinew Ali Ayele, Elena Tutubalina, Aung Kyaw Htet, Xintong Wang, Surendrabikram Thapa, Tanmoy Chakra… — SemEval-2026 Task 9: Detecting Multilingual, Multicultural and Multievent Online Polarization
  14. Usman Naseem, Robert Geislinger, Juan Ren, Sarah Kohail, Rudy Garrido Veliz, P Sam Sahil, Yiran Zhang, Marco Antonio Stranisci, Idris Abdulmumin, Özge Alacam, Cengiz Acarürk, Aisha Jabr, Saba Anwar, Abinew Ali Ayele, Simona Frenda, Alessandra Teresa Cignarella, Elena Tutubalina, Oleg Rogov, Aung Kya… — POLAR: A Benchmark for Multilingual, Multicultural, and Multi-Event Online Polarization
  15. Markus Prior. 2013. Media and political polarization. Annual Review of Political Science, 16:101–127. find
  16. Tal Ridnik, Emanuel Ben-Baruch, Niv Zamir, Asaf Noy, Itamar Friedman, Matan Protter, and Lihi Zelnik-Manor. 2021. Asymmetric loss for multi-label classification. In Proceedings of ICCV, pages 82–91. find
  17. Sara Rosenthal, Noura Farra, and Preslav Nakov. 2017. SemEval-2017 Task 4: Sentiment analysis in Twitter. In Proceedings of SemEval-2017, pages 502–518. averages them equally across all classes, as outlined in Eq. (B.1): C Cass R. Sunstein. 2002. The law of group polarization. Journal of Political P…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mahmud-2026-argonauts-semeval-2026-task-9/