Elamur Research⌕
☰
home / papers / 2026 / UIT-Polar at SemEval-2026 Task 9 Detecting Multilingual, Multicultural and Multievent Online Polarization

UIT-Polar at SemEval-2026 Task 9 Detecting Multilingual, Multicultural and Multievent Online Polarization

Paper · 2026 · Hoàn Trần

Summary

This paper describes the UIT-Polar submission to SemEval-2026 Task 9, Subtask 1. The task is to decide whether a social media post expresses attitude polarization, framed as binary classification. The authors worked only on the English track, where non-polarized posts heavily outnumber polarized ones. Their pipeline has two stages and works from coarse to fine.

  • Stage 1: a fine-tuned DeBERTa-v3-large classifier labels the posts it is confident about.
  • Stage 2: low-confidence posts are routed to Mistral-7B-Instruct-v0.2. This model is adapted with LoRA and prompted with retrieved few-shot examples, balanced between the two classes.

On the official test set the system reached 81.62 Macro-F1, beating the 78.02 baseline and finishing fifth. On the development set the combined system outperformed either component used alone. However, dev scores were notably higher than test scores. The authors attribute this gap to the confidence threshold, which was hard to tune with only five test submissions allowed. The work shows a practical way to combine a cheap encoder with an LLM so that expensive reasoning is reserved for borderline cases in imbalanced classification.

Key points

  • Routing: posts whose maximum class probability is at least τ = 0.8 keep the DeBERTa label. On the dev set this covered about 70% of posts, and the remaining ~30% went to Mistral.
  • Stage 2 setup: LoRA uses r = 32 and α = 64, applied to the attention projections and gate_proj. Few-shot examples are retrieved with a Qwen embedding model via cosine similarity and ordered alternately positive/negative. Three pairs (3+3) worked best.
  • Dev-set ablation (Macro-F1):
    • Mistral zero-shot: 80.61
    • Mistral few-shot 3+3: 82.54
    • Mistral with LoRA plus few-shot: 84.83
  • Dev-set component comparison (Macro-F1): DeBERTa alone scored 79.64, the LLM alone 84.83, and the full two-stage system 86.57.
  • Official result: 81.62 Macro-F1 on the English test set versus a 78.02 baseline, placing 5th. The authors identify sensitivity to τ as the main limitation.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

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

Cite

Cite as Hoàn Trần. 2026. UIT-Polar at SemEval-2026 Task 9 Detecting Multilingual, Multicultural and Multievent Online Polarization. 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-1.19

@inproceedings{tran2026uit,
  title = {{UIT-Polar at SemEval-2026 Task 9 Detecting Multilingual, Multicultural and Multievent Online Polarization}},
  author = {Hoàn Trần},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.19},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.19}
}

References

11 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Alex Castro-Ros, Marie Pellat, Kevin Robinson, and 16 others. 2024.… find
  2. Pengcheng He, Jianfeng Gao, and Weizhu Chen. 2023. Debertav3: Improving deberta using electra-style pre-training with gradient-disentangled embedding sharing. In International Conference on Learning Representations (ICLR).
  3. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F Christiano, Jan Leike, and Ryan Lowe. 2022. Tra… — Training language models to follow instructions with human feedback
  4. Jason Wei, Maarten Bosma, Vincent Y. Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M. Dai, and Quoc V. Le. 2022. Finetuned language models are zero-shot learners. Preprint, arXiv:2109.01652.
  5. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. Preprint, arXiv:2106.09685.
  6. Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou. 2025. Qwen3 embedding: Advancing text embedding and reranking through foundation models. Preprint, arXiv:2506.05176.
  7. Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix,… find
  8. Zihao Zhao, Eric Wallace, Shi Feng, Dan Klein, and Sameer Singh. 2021. Calibrate before use: Improving few-shot performance of language models. In Proceedings of the 38th International Conference on Machine Learning, volume 139 of Proceedings of Machine Learning Research, pages 12697–12706. PMLR.
  9. Jiachang Liu, Dinghan Shen, Yizhe Zhang, Bill Dolan, Lawrence Carin, and Weizhu Chen. 2022. What makes good in-context examples for GPT-3? In Proceedings of Deep Learning Inside Out (DeeLIO 2022): The 3rd Workshop on Knowledge Extraction and Integration for Deep Learning Architectures, pages 100–114…
  10. Usman Naseem, Robert Geislinger, Juan Ren, Sarah Kohail, Rudy Garrido Veliz, P Sam Sahil, Yiran Zhang, Marco Antonio Stranisci, Idris Abdulmumin, "Ozge Alacam, Cengiz Acart"urk, Aisha Jabr, Saba Anwar, Abinew Ali Ayele, Elena Tutubalina, Aung Kyaw Htet, Xintong Wang, Surendrabikram Thapa, Tanmoy Cha… — SemEval-2026 Task 9: Detecting Multilingual, Multicultural and Multievent Online Polarization
  11. Usman Naseem, Robert Geislinger, Juan Ren, Sarah Kohail, Rudy Garrido Veliz, P Sam Sahil, Yiran Zhang, Marco Antonio Stranisci, Idris Abdulmumin, Özge Alacam, Cengiz Acartürk, Aisha Jabr, Saba Anwar, Abinew Ali Ayele, Simona Frenda, Alessandra Teresa Cignarella, Elena Tutubalina, Oleg Rogov, Aung Ky… — POLAR: A Benchmark for Multilingual, Multicultural, and Multi-Event Online Polarization

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tran-2026-uit-polar-semeval-2026-task/