Elamur Research⌕
☰
home / papers / 2026 / CYUT at SemEval-2026 Task 3: Multi-Task Dimensional Aspect Sentiment Regression with Polar Multi-Zone Labeling in VA Space

CYUT at SemEval-2026 Task 3: Multi-Task Dimensional Aspect Sentiment Regression with Polar Multi-Zone Labeling in VA Space

Paper · 2026 · Shih-Hung Wu, Xian-Yan Chen and Yi-Min Jian

Summary

The paper presents CYUT's entry for SemEval-2026 Task 3 Track B. The task asks systems to predict continuous Valence–Arousal (VA) scores, each in the range [1, 9], for a text and a target aspect across several languages. Systems are scored by RMSE. The authors split each text into one instance per aspect. The main task is VA regression, trained alongside auxiliary classification tasks generated automatically from the gold scores: polarity, intensity and VA quadrant.

Fixed-threshold auxiliary labels can leave some areas of the VA plane with very few examples. To address this, the authors add Polar Multi-Zone Labeling (PMZL-7). It treats the VA plane in polar coordinates around (5, 5). Samples closest to the centre form a neutral core, and the remaining samples are split by angle into six zones holding roughly equal numbers of examples.

They test XLM-R, Qwen2 and Ministral. PMZL-7 helps the two generative models but not XLM-R. The Qwen2 system beats both official baselines and achieves the top score among all teams on Nigerian Pidgin. The work suggests that auxiliary labels which account for the data distribution are worth comparing, although their benefit depends on the backbone.

Key points

  • The auxiliary labels are defined as follows:
    • Polarity: positive if V ≥ 5.5, negative if V ≤ 4.5, otherwise neutral.
    • Intensity: excited if A ≥ 6.0, calm if A ≤ 4.0, otherwise medium.
    • Quadrant: split at the point (5, 5).
    • PMZL-7: the 20% of samples nearest the centre become the neutral zone, and the rest fall into six equal-mass angular zones.
  • The backbones are XLM-RoBERTa-base, Qwen2-7B and Ministral-7B. The generative models produce outputs in a "V#A" format, with regex recovery and clipping to [1, 9] when parsing fails.
  • Official test average RMSE is 1.3999 for Qwen2-7B, 1.4817 for Ministral-7B and 1.6612 for XLM-R. The baselines score 2.4604 (mBERT) and 1.6024 (Mistral-3 14B). Qwen2 reaches 1.1024 on PCM (Nigerian Pidgin), the best result in the competition.
  • Development-set ablations cover ENG, DEU, ZHO, PCM and SWA:
    • Adding PMZL-7 lowers average RMSE for Qwen2 (1.5462 → 1.5184) and Ministral (1.4372 → 1.4154).
    • For XLM-R, the model performs slightly better without PMZL-7 (1.5615 vs 1.5820).
    • Dropping intensity supervision hurts all three backbones.
  • Code is released at github.com/YiMin0130/CYUT-SemEval2026-Task3. Planned future work includes other zone counts and soft zone assignment.

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

🎓Chaoyang University of TechnologyTaichung · Taiwan2 works · 2 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksPsychology · 1,649 worksExperimental and Cognitive Psychology · 930 works

🧭 Topics 1Research subjects and languages the work is about.Sentiment & emotion · 1,035 works

Cite

Cite as Shih-Hung Wu, Xian-Yan Chen and Yi-Min Jian. 2026. CYUT at SemEval-2026 Task 3: Multi-Task Dimensional Aspect Sentiment Regression with Polar Multi-Zone Labeling in VA Space. 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-1.23

@inproceedings{wu2026cyut,
  title = {{CYUT at SemEval-2026 Task 3: Multi-Task Dimensional Aspect Sentiment Regression with Polar Multi-Zone Labeling in VA Space}},
  author = {Shih-Hung Wu, Xian-Yan Chen and Yi-Min Jian},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.23},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.23}
}

References

21 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jonas Becker, Liang-Chih Yu, Shamsuddeen Hassan Muhammad, Jan Philip Wahle, Terry Ruas, Idris Abdulmumin, Lung-Hao Lee, Nelson Odhiambo, Lilian Wanzare, Wen-Ni Liu, Tzu-Mi Lin, Zhe-Yu Xu, Ying-Lung Lin, Jin Wang, Maryam Ibrahim Mukhtar, Bela Gipp, and Saif M. Mohammad. 2026. Dimstance: Multilingual… find
  2. Sven Buechel, Anneke Buffone, Barry Slaff, Lyle Ungar, and João Sedoc. 2018. Modeling empathy and distress in reaction to news stories. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 4758–4765, Brussels, Belgium. Association for Computational Linguis…
  3. Sven Buechel and Udo Hahn. 2017. Emobank: Studying the impact of annotation perspective and representation format on dimensional emotion analysis. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics. Association for Computational Linguistics… find
  4. Sven Buechel and Udo Hahn. 2018. Word emotion induction for multiple languages as a deep multi-task learning problem. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1907–1918. Association for Co… find
  5. Rich Caruana. 1997. Multitask learning. Machine Learning, 28(1):41–75. find
  6. 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. In Proceedings of the 58th Annual Meeting of the Association fo…
  7. Alan S. Cowen and Dacher Keltner. 2017. Self-report captures 27 distinct categories of emotion bridged by continuous gradients. Proceedings of the National Academy of Sciences, 114(38):E7900–E7909. find
  8. 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. arXiv preprint arXiv:2106.09685.
  9. Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, György Lengyel, Guillaume Lample, Luc Saulnier, and 1 others. 2023. Mistral 7b. arXiv preprint arXiv:2310.06825. find
  10. Dilek Küçük and Fazli Can. 2020. Stance detection: A survey. ACM Computing Surveys, 53(1). find
  11. Lung-Hao Lee, Jheng-Hong Li, and Liang-Chih Yu. 2022. Chinese EmoBank: Building valence-arousal resources for dimensional sentiment analysis. ACM Transactions on Asian and Low-Resource Language Information Processing, 21(4):1–19. find
  12. Saif M. Mohammad, Parinaz Sobhani, and Svetlana Kiritchenko. 2017. Stance and sentiment in tweets. ACM Transactions on Internet Technology, 17(3):1– 23.
  13. Rajdeep Mukherjee, Aakash Naik, Saptarshi Poddar, Somak Dasgupta, and Niloy Ganguly. 2021. Understanding the role of affect dimensions in detecting emotions from tweets: A multi-task approach. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Re… find
  14. Sungjoon Park, Jiseon Kim, Seonghyeon Ye, Jaeyeol Jeon, Hee Young Park, and Alice Oh. 2021. Dimensional emotion detection from categorical emotion. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 4367–4380. Association for Computational Linguistics. find
  15. James A. Russell. 1980. A circumplex model of affect. Journal of Personality and Social Psychology, 39(6):1161–1178. find
  16. Harold Schlosberg. 1954. Three dimensions of emotion. Psychological Review, 61(2):81–88. find
  17. Hongfei Xie, Eduard Hovy, and Diyi Yang. 2021. A multi-dimensional relation model for dimension score prediction. Information Sciences, 579:832– 844. find
  18. An Yang and 1 others. 2024. Qwen2 technical report. arXiv preprint arXiv:2407.10671. find
  19. Liang-Chih Yu, Jonas Becker, Shamsuddeen Hassan Muhammad, Idris Abdulmumin, Lung-Hao Lee, Ying-Lung Lin, Jin Wang, Jan Philip Wahle, Terry Ruas, Alexander Panchenko, Ilseyar Alimova, Kai-Wei Chang, Lilian Wanzare, Nelson Odhiambo, Bela Gipp, and Saif M. Mohammad. 2026. SemEval-2026 task 3: Dimension… find
  20. Liang-Chih Yu, Lung-Hao Lee, Jin Wang, and Kam-Fai Wong. 2017. IJCNLP-2017 task 2: Dimensional sentiment analysis for chinese phrases. In Proceedings of the IJCNLP 2017 Shared Tasks, pages 17–24. Asian Federation of Natural Language Processing. find
  21. Siyi Zhu, Yu Zhang, and Xiaobing Li. 2019. Adversarial attention modeling for multi-dimensional emotion regression. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 631–640. Association for Computational Linguistics. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wu-2026-cyut-semeval-2026-task-3/