Elamur Research⌕
☰
home / papers / 2026 / OZemi at SemEval-2026 Task 9: A Cross-Lingual Approach to Online Text Polarization Classification Using Multilingual Models and Adaptive Loss Formulation

OZemi at SemEval-2026 Task 9: A Cross-Lingual Approach to Online Text Polarization Classification Using Multilingual Models and Adaptive Loss Formulation

Paper · 2026 · Hidetsune Takahashi, Eleale Nusi Tee, Aika Yu et al.

Summary

The paper describes the Waseda University team's (OZemi) system for SemEval-2026 Task 9 on detecting polarization in online text. The task has three subtasks. Subtask 1 is binary detection of polarized posts in 22 languages. Subtask 2 is multi-label classification of polarization type (5 labels, 22 languages). Subtask 3 is multi-label classification of how polarization is expressed (6 labels, 18 languages). The data are strongly skewed, and some language–label combinations are extremely sparse.

Rather than tuning each language separately, the authors built one shared pipeline. It fine-tunes TwHIN-BERT-base, a multilingual encoder pre-trained on Twitter, using a cross-entropy loss whose class weights grow as class frequency shrinks. Back-translation was added only where it helped. The paper also includes a surface-level stylistic analysis and an error analysis.

In Subtask 1, macro F1 exceeded 70% for 20 of the 22 languages, and the system ranked first for Persian. The work shows that a single, fairly simple multilingual setup can stay competitive across many languages, including low-resource ones.

Key points

  • Base model: TwHIN-BERT-base was chosen over RoBERTa-base and XLM-RoBERTa-base. Its largest gain was on Italian, where development macro F1 rose from 0.352 to 0.606. For English, however, the monolingual RoBERTa-base remained stronger (0.828 vs 0.704).
  • Loss function: class weights are set to w_k = N/N_k. For very rare labels this produces extreme values, such as roughly 405.7 for Hausa "invalidation" (9 positives out of 3,651), which the authors note may destabilize training.
  • Back-translation: the authors used MarianMT (Helsinki-NLP opus-mt) models to back-translate training data, and the effect varied by language:
    • Hindi improved from 0.7475 to 0.7749 with XLM-RoBERTa-base.
    • English dropped from 0.85 to 0.8058, and Spanish and Russian also got worse.
    • The Indian-language model MuRIL reached only 0.72 on Hindi.
  • Rankings:
    • Subtask 1: 1st of 44 for Persian and 3rd of 45 for Hausa.
    • Subtask 2: 5th of 27 for Swahili.
    • Subtask 3: 2nd of 19 for both Persian and Hausa.
  • Error analysis: the Italian model tends to treat emotional, expressive style as polarization (false positives). The German model misses polarization conveyed through irony or argumentative framing (false negatives).

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

🎓Waseda UniversityTokyo · Japan20 works · 584 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 Hidetsune Takahashi, Eleale Nusi Tee, Aika Yu, Ruri Furukawa, Sooeun Kim, Shuta Niinomi, Dingyu Zhang and Emily Ohman. 2026. OZemi at SemEval-2026 Task 9: A Cross-Lingual Approach to Online Text Polarization Classification Using Multilingual Models and Adaptive Loss Formulation. 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-1.27

@inproceedings{takahashi2026ozemi,
  title = {{OZemi at SemEval-2026 Task 9: A Cross-Lingual Approach to Online Text Polarization Classification Using Multilingual Models and Adaptive Loss Formulation}},
  author = {Hidetsune Takahashi, Eleale Nusi Tee, Aika Yu, Ruri Furukawa, Sooeun Kim, Shuta Niinomi, Dingyu Zhang and Emily Ohman},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.27},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.27}
}

References

16 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Swapan Deep Arora, Guninder Pal Singh, Anirban Chakraborty, and Moutusy Maity. 2022. Polarization and social media: A systematic review and research agenda. Technological Forecasting and Social Change, 183:121942. find
  2. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association for Computational Li… find
  3. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies,…
  4. Sergey Edunov, Myle Ott, Michael Auli, and David Grangier. 2018. Understanding back-translation at scale. In Proceedings of the 2018 conference on empirical methods in natural language processing, pages 489–500. find
  5. Marcin Junczys-Dowmunt, Roman Grundkiewicz, Tomasz Dwojak, Hieu Hoang, Kenneth Heafield, Tom Neckermann, Frank Seide, Ulrich Germann, Alham Fikri Aji, Nikolay Bogoychev, and 1 others. 2018. Marian: Fast neural machine translation in c++. In Proceedings of ACL 2018, system demonstrations, pages 116–1…
  6. Simran Khanuja, Diksha Bansal, Sarvesh Mehtani, Savya Khosla, Atreyee Dey, Balaji Gopalan, Dilip Kumar Margam, Pooja Aggarwal, Rajiv Teja Nagipogu, Shachi Dave, and 1 others. 2021. Muril: Multilingual representations for indian languages. arXiv preprint arXiv:2103.10730.
  7. 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.
  8. 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
  9. Hidetsune Takahashi, Sumiko Teng, Jina Lee, Wenxiao Hu, Rio Obe, Chuen Shin Yong, and Emily Ohman. 2025. OZemi at SemEval-2025 task 11: Multilingual emotion detection and intensity. In Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), pages 109–115, Vienna, Austri… find
  10. Jörg Tiedemann and Santhosh Thottingal. 2020. OPUS-MT – building open translation services for the world. In Proceedings of the 22nd Annual Conference of the European Association for Machine Translation, pages 321–322, Lisboa, Portugal. find
  11. Xinyang Zhang, Yury Malkov, Omar Florez, Serim Park, Brian McWilliams, Jiawei Han, and Ahmed El-Kishky. 2023. Twhin-bert: A socially-enriched pre-trained language model for multilingual tweet representations at twitter. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data… find
  12. 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
  13. Marc Pàmies, Emily Öhman, Kaisla Kajava, and Jörg Tiedemann. 2020. Lt@ helsinki at semeval-2020 task 12: Multilingual or language-specific bert? In Proceedings of the Fourteenth Workshop on Semantic Evaluation, pages 1569–1575. find
  14. Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Improving neural machine translation models with monolingual data. In Proceedings of the 54th annual meeting of the association for computational linguistics (volume 1: long papers), pages 86–96. find
  15. Hidetsune Takahashi, Xingru Lu, Sean Ishijima, De-okgyu Seo, Yongju Kim, Sehoon Park, Min Song, Kathylene Marante, Keitaro-luke Iso, Hirotaka Tokura, and Emily Ohman. 2024. OZemi at SemEval-2024 task 1: A simplistic approach to textual relatedness evaluation using transformers and machine translatio… find
  16. Amharic Arabic Bengali Chinese English German Hausa Hindi Khmer Nepali Odia Punjabi Spanish Swahili Telugu Turkish Urdu Persian *Table 7: Subtask 3 positive label proportions (%) per language. Multi-label; rows do not sum to 100%.* Language Emoji % Emoji Density @Mention % \\\#Hashtag % Repeat Punct.… find

Updated

↑ back to top

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