Elamur Research⌕
☰
home / papers / 2026 / NCL-BU at SemEval-2026 Task 3: Fine-tuning XLM-RoBERTa for Multilingual Dimensional Sentiment Regression

NCL-BU at SemEval-2026 Task 3: Fine-tuning XLM-RoBERTa for Multilingual Dimensional Sentiment Regression

Paper · 2026 · Tong Wu, Nicolay Rusnachenko and Huizhi(elly) Liang

Summary

This paper reports a system for SemEval-2026 Task 3 (DimABSA), Track A, Subtask 1. In this subtask, each aspect mentioned in a sentence receives two continuous scores, valence and arousal, on a 1–9 scale, rather than a categorical polarity label. The authors fine-tune XLM-RoBERTa-base on input pairs of the form "[CLS] text [SEP] aspect [SEP]". Two separate two-layer MLP heads read the [CLS] vector, and a scaled sigmoid keeps each output in [1, 9]. The loss is the sum of the valence and arousal MSEs. One model is trained per language–domain pair: English restaurant and laptop, and Chinese restaurant, laptop and finance. For the final submission, the training and development data are merged.

The fine-tuned encoder beats both organizer baselines on every dataset. It also outperforms few-shot prompting of large LLMs. Error analysis shows two weaknesses: the model handles strongly negative aspects poorly, and it over-smooths when several aspects in one sentence differ in sentiment. The work suggests that a compact, task-specific multilingual encoder remains a strong choice for affective regression compared with prompted LLMs.

Key points

  • Test RMSE_VA (lower is better) was 1.4562 for eng_lap, 1.4861 for eng_res, 0.7510 for zho_lap, 0.9553 for zho_res and 0.5391 for zho_fin. Relative to the stronger of the two baselines, Kimi-K2 Thinking and QLoRA-tuned Qwen-3 14B, the gains were 31–34% on English and 50–63% on Chinese.
  • On an 80/20 split of the training data, the system was compared with few-shot prompting (6 examples, temperature 0.1) of GPT-5.2, LLaMA-3-70B-Instruct and LLaMA-4-Maverick-Instruct. The fine-tuned model scored better on every pair.
  • Training settings: max length 256, batch size 16, learning rate 2e-5 with AdamW, 10% warmup, at most 10 epochs, early-stopping patience of 3, and a single run with seed 42. The data come from the DimABSA dataset, and the English portion builds on ACOS.
  • Training labels are skewed toward positive values, especially in English. As a result, errors are highest for aspects whose gold valence falls below 3, where RMSE_VA reaches 2.43 in the restaurant domain and 2.20 in the laptop domain.
  • Valence was harder to predict than arousal in four of the five datasets; the largest gap was on English laptop (1.25 vs 0.75). The authors propose aspect-aware pooling and data augmentation as future fixes.

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

🎓Newcastle UniversityNewcastle upon Tyne · United Kingdom19 works · 746 citations🎓Bournemouth UniversityPoole · United Kingdom5 works · 276 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

🧭 Topics 2Research subjects and languages the work is about.Sentiment & emotion · 1,035 worksMultilingual & cross-lingual NLP · 882 works

Cite

Cite as Tong Wu, Nicolay Rusnachenko and Huizhi(elly) Liang. 2026. NCL-BU at SemEval-2026 Task 3: Fine-tuning XLM-RoBERTa for Multilingual Dimensional Sentiment Regression. 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-1.240

@inproceedings{wu2026ncl,
  title = {{NCL-BU at SemEval-2026 Task 3: Fine-tuning XLM-RoBERTa for Multilingual Dimensional Sentiment Regression}},
  author = {Tong Wu, Nicolay Rusnachenko and Huizhi(elly) Liang},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.240},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.240}
}

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. 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: Volume 2, Short Papers, pages 578–585, Va… find
  2. Hongjie Cai, Rui Xia, and Jianfei Yu. 2021. Aspect-category-opinion-sentiment quadruple extraction with implicit aspects and opinions. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing…
  3. Alexis Conneau, Karttikeya Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Marius Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Associatio…
  4. 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,…
  5. Lung-Hao Lee, Jian-Hong Li, and Liang-Chih Yu. 2022. Chinese emobank: Building valence-arousal resources for dimensional sentiment analysis. ACM Trans. Asian Low-Resour. Lang. Inf. Process., 21(4). find
  6. 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
  7. 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 10th SIGHAN Workshop on Chinese Language Processing (SIGHAN-10), pages 165–174, Bangkok, Thailand. Association for Compu… find
  8. Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. In International Conference on Learning Representations. find
  9. 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
  10. Maria Pontiki, Dimitris Galanis, Haris Papageorgiou, Ion Androutsopoulos, Suresh Manandhar, Mohammad AL-Smadi, Mahmoud Al-Ayyoub, Yanyan Zhao, Bing Qin, Orphée De Clercq, Véronique Hoste, Marianna Apidianaki, Xavier Tannier, Natalia Loukachevitch, Evgeniy Kotelnikov, Nuria Bel, Salud María Jiménez-Z… — SemEval-2016 Task 5: Aspect Based Sentiment Analysis
  11. Maria Pontiki, Dimitris Galanis, Haris Papageorgiou, Suresh Manandhar, and Ion Androutsopoulos. 2015. SemEval-2015 task 12: Aspect based sentiment analysis. In Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 486–495, Denver, Colorado. Association for Comput…
  12. Maria Pontiki, Dimitris Galanis, John Pavlopoulos, Harris Papageorgiou, Ion Androutsopoulos, and Suresh 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. Associa…
  13. James A Russell. 1980. A circumplex model of affect. Journal of Personality and Social Psychology, 39(6):1161–1178. find
  14. 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
  15. Chen Zhang, Qiuchi Li, Dawei Song, and Linqi Song. 2021. Aspect sentiment quad prediction as paraphrase generation. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9209–9219.
  16. Wenxuan Zhang, Yue Deng, Bing Liu, Sinno Pan, and Lidong Bing. 2024. Sentiment analysis in the era of large language models: A reality check. In Findings of the Association for Computational Linguistics: NAACL 2024, pages 3881–3906, Mexico City, Mexico. Association for Computational Linguistics.
  17. 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

Updated

↑ back to top

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