Elamur Research⌕
☰
home / papers / 2026 / SCUZANE at SemEval-2026 Task 3: Dimensional Aspect-based Sentiment Analysis with Supervised Contrastive Regression and R-Drop Regularization

SCUZANE at SemEval-2026 Task 3: Dimensional Aspect-based Sentiment Analysis with Supervised Contrastive Regression and R-Drop Regularization

Paper · 2026 · Ziang Zhou, Xiangmei He and Chenhongyi Bai

Summary

The paper describes team SCUZANE's entry for SemEval-2026 Task 3 (DimABSA), Track A, Subtask 1 (DimASR). Instead of assigning discrete polarity labels, systems must predict real-valued valence and arousal (VA) scores between 1 and 9 for each aspect mentioned in a text. Data cover Restaurant, Laptop, Hotel and Finance domains in six languages, including Ukrainian.

The authors frame the task as sentence-pair regression. Each aspect becomes a short prompt placed before the text, and DeBERTa-v3-large encodes the result. Several additions target the instability and noise typical of regression:

  • [CLS] vectors from the final four layers are pooled together.
  • Multi-sample dropout feeds a shared regression head.
  • R-Drop enforces consistent predictions under different dropout masks.
  • A contrastive loss weights pairs by how close their VA labels are.
  • The regression loss mixes MSE with a Concordance Correlation Coefficient (CCC) term.

The system lands mostly mid-table, with its best placings on Chinese. It beats the LLM baselines on nearly every dataset. This makes it a practical reference point for fine-grained, multilingual sentiment work, including on Ukrainian.

Key points

  • Ukrainian restaurant results: RMSE 1.5730, PCC-V 0.8891, PCC-A 0.5930, ranked 11 of 20. Arousal correlation is clearly weaker than valence in every language.
  • Best and worst datasets: The best was zho-finance (RMSE 0.5117, rank 7/21). The weakest was tat-restaurant (RMSE 2.3199, rank 19/22).
  • Comparison with baselines: SCUZANE achieves lower RMSE than the Kimi-K2 Thinking and Qwen-3 14B baselines everywhere except Tatar. There, Kimi-K2 scores better (1.938).
  • Training objective:
    • Regression loss = 0.3·MSE + 0.7·CCC.
    • Total loss adds R-Drop with weight α = 1.0 and the label-distance-weighted contrastive term with weight β = 0.2.
    • Contrastive pair weights are exp(−‖yᵢ−yⱼ‖²).
  • Training setup and gaps:
    • Max length 128 tokens, AdamW optimizer.
    • Learning rates of 1e-5 for the backbone and 5e-5 for the heads.
    • Five dropout masks with p = 0.2.
    • Early stopping on average PCC with patience 15.
    • The paper includes no ablation study.

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

🎓Sichuan UniversityChengdu · China20 works · 1,293 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 Ziang Zhou, Xiangmei He and Chenhongyi Bai. 2026. SCUZANE at SemEval-2026 Task 3: Dimensional Aspect-based Sentiment Analysis with Supervised Contrastive Regression and R-Drop Regularization. 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-1.251

@inproceedings{zhou2026scuzane,
  title = {{SCUZANE at SemEval-2026 Task 3: Dimensional Aspect-based Sentiment Analysis with Supervised Contrastive Regression and R-Drop Regularization}},
  author = {Ziang Zhou, Xiangmei He and Chenhongyi Bai},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.251},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.251}
}

References

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

  1. Pengcheng He, Jianfeng Gao, and Weizhu Chen. 2021. Debertav3: Improving deberta using electra-style pre-training with gradient-disentangled embedding sharing. Preprint, arXiv:2111.09543.
  2. Lung-Hao Lee, Jian-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):1–18. find
  3. Lung-Hao Lee, Liang-Chih Yu, Natalia Loukashe-vich, Ilseyar Alimova, Alexander Panchenko, Tzu-Mi Lin, Zhe-Yu Xu, Jian-Yu Zhou, Guangmin Zheng, Jin Wang, Sharanya Awasthi, Jonas Becker, Jan Philip Wahle, Terry Ruas, Shamsuddeen Hassan Muhammad, and Saif M. Mohammad. 2026. Dimabsa: Building multilingu… find
  4. Lung-Hao Lee, Liang-Chih Yu, Suge Wang, and Jian Liao. 2024. Overview of the sighan 2024 shared task for chinese dimensional aspect-based sentiment analysis. In Proceedings of the 12th Workshop on Chinese Language Processing (SIGHAN-12), pages 180–189, Miami, Florida. Association for Computational L… find
  5. Manandhar. 2014. SemEval-2014 task 4: Aspect based sentiment analysis. In Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 27–35, Dublin, Ireland. Association for Computational Linguistics.
  6. James A. Russell. 1980. A circumplex model of affect. Journal of Personality and Social Psychology, 39(6):1161–1178. find
  7. James A. Russell. 2003. Core affect and the psychological construction of emotion. Psychological Review, 110(1):145–172. find
  8. 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
  9. Wenxuan Zhang, Yang Deng, Xin Li, Yifei Yuan, Lidong Bing, and Wai Lam. 2021. Aspect sentiment quad prediction as paraphrase generation. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9209– 9219. Association for Computational Linguistics.
  10. Wenxuan Zhang, Xin Li, Yang Deng, Lidong Bing, and Wai Lam. 2023. A survey on aspect-based sentiment analysis: Tasks, methods, and challenges. IEEE Transactions on Knowledge and Data Engineering, 35(11):11019–11038. find
  11. Xiaobo Liang, Lijun Wu, Juntao Li, Yue Wang, Qi Meng, Tao Qin, Wei Chen, Min Zhang, and Tie-Yan Liu. 2021. R-drop: Regularized dropout for neural networks. Advances in Neural Information Processing Systems, 34:10890–10905. find
  12. Saif M. Mohammad. 2018. Obtaining reliable human ratings of valence, arousal, and dominance for 20,000 english words. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 174–184, Melbourne, Australia. Association for Computational…
  13. Haiyun Peng, Lu Xu, Lidong Bing, Fei Huang, Wei Lu, and Luo Si. 2020. Knowing what, how and why: A near complete solution for aspect-based sentiment analysis. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 8600–8607. find
  14. Maria Pontiki, Dimitris Galanis, John Pavlopoulos, Harris Papageorgiou, Ion Androutsopoulos, and Suresh find

Updated

↑ back to top

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