Elamur Research⌕
☰
home / papers / 2026 / QuadAI at SemEval-2026 Task 3: Ensemble Learning of Hybrid RoBERTa and LLMs for Dimensional Aspect-Based Sentiment Analysis

QuadAI at SemEval-2026 Task 3: Ensemble Learning of Hybrid RoBERTa and LLMs for Dimensional Aspect-Based Sentiment Analysis

Paper · 2026 · A.J.W. de Vink, Filippos Karolos Ventirozos, Natalia Amat-Lefort et al.

Summary

The paper describes the QuadAI system for SemEval-2026 Task 3, subtask A.1 (DimASR). The task asks for a valence and an arousal score, each on a 1–9 scale, for every given aspect in a review. The authors also ran development experiments on Track-B (DimStance).

Their main encoder is a hybrid RoBERTa model with two heads trained jointly. One head regresses the scores directly. The other splits the target range into bins, classifies with cross-entropy and converts the result to an expected value. The two outputs are then averaged.

They also prompt LLMs with in-context examples chosen by embedding similarity. Before prompting, they filter noisy labels: training items are clustered with HDBSCAN in VA space, and three LLMs vote on which items look mislabelled.

Encoder and LLM predictions are merged by averaging, weighted averaging or ridge stacking. On the laptop dev set, this ensemble clearly beat each individual model. Only the hybrid RoBERTa reached the official test submission, where it placed mid-table. The work suggests that encoder and LLM predictions complement each other for continuous sentiment regression.

Key points

  • Hybrid RoBERTa on restaurant dev: MSE fell to 0.4919, compared with 0.8176 for the regression-only head. The best configuration used 11 bins, alpha_cls = 0.2 and a 0.5 averaging weight.
  • LLM setup: in-context examples were retrieved with OpenAI text-embedding-3-large. Gemini, Claude and GPT-5.2 removed an example from the candidate pool only when all three flagged it. Gemini was then used for the final prompting.
  • Ensemble on laptop dev: weighted averaging (weights [0.3, 0.7]) and out-of-fold ridge stacking both reached RMSE 0.6344. This compares with 0.695 for the LLM alone and about 0.736 for the best hybrid encoder. Adding VADER features gave no gain.
  • Official A.1 results (hybrid RoBERTa only):
    • Laptop: 16th of 30, with a score of 1.4062 (best team 1.2408, baseline 2.8053).
    • Restaurant: 22nd of 33, with a score of 1.3632 (best team 1.1035, baseline 2.791).
  • Track-B dev (English, Environmental Protection): the hybrid model with 21 bins and w = 0.3 reached RMSE 1.4022, slightly better than either head alone. No Track-B run was submitted.

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

🎓Leiden UniversityLeiden · Netherlands32 works · 1,074 citations🎓Manchester Metropolitan UniversityManchester · United Kingdom5 works · 205 citations🤝Leiden University Medical CenterLeiden · Netherlands4 works · 2 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksPsychology · 1,552 worksExperimental and Cognitive Psychology · 870 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,320 worksSentiment & emotion · 972 works

Cite

Cite as A.J.W. de Vink, Filippos Karolos Ventirozos, Natalia Amat-Lefort and Lifeng Han. 2026. QuadAI at SemEval-2026 Task 3: Ensemble Learning of Hybrid RoBERTa and LLMs for Dimensional Aspect-Based Sentiment Analysis. 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-1.11

@inproceedings{vink2026quadai,
  title = {{QuadAI at SemEval-2026 Task 3: Ensemble Learning of Hybrid RoBERTa and LLMs for Dimensional Aspect-Based Sentiment Analysis}},
  author = {A.J.W. {de Vink}, Filippos Karolos Ventirozos, Natalia Amat-Lefort and Lifeng Han},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.11},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.11}
}

References

21 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Dataset Rank RMSE Baseline Laptop Restaurant 16 / 29 22 / 32 1.4062 1.3632 2.8053 2.7910 find
  2. Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. 2019. Optuna: A next-generation hyperparameter optimization framework. Preprint, arXiv:1907.10902. *Table 7: Summary of our system performance on the SemEval-2026 Track A.1 datasets. Lower RMSE is better.* Jonas Becker,…
  3. Souvic Chakraborty, Pawan Goyal, and Animesh Mukherjee. 2020. Aspect-based sentiment analysis of scientific reviews. In Proceedings of the ACM/IEEE Joint Conference on Digital Libraries in 2020, pages 207–216. find
  4. Amit Chauhan, Aman Sharma, and Rajni Mohana. 2025. An enhanced aspect-based sentiment analysis model based on roberta for text sentiment analysis. Informatica, 49(14). stacked and voted ensembles on LLMs. In Proceedings of the Second Workshop on Patient-Oriented Language Processing (CL4Health), page… find
  5. AJW de Vink, Natalia Amat-Lefort, and Lifeng Han. 2025. Reviewgraph: A knowledge graph embedding based framework for review rating prediction with sentiment features. arXiv preprint arXiv:2508.13953. — [arXiv:2508.13953](https://arxiv.org/abs/2508.13953) find
  6. Xuefeng Shi, Weiping Ding, Min Hu, Xin Kang, and Fuji Ren. 2025. Triple dimensional psychology knowledge encouraging graph attention networks to exploit aspect-based sentiment analysis. Scientific Reports, 15(1):27109. find
  7. C. J. Hutto and Eric Gilbert. 2014. Vader: A parsimonious rule-based model for sentiment analysis of social media text. In Proceedings of the 8th International Conference on Weblogs and Social Media (ICWSM). find
  8. Filippos Ventirozos, Peter A. Appleby, and Matthew Shardlow. 2025a. Are you sure you’re positive? consolidating chain-of-thought agents with uncertainty quantification for aspect-category sentiment analysis. In Proceedings of the 1st Workshop for Research on Agent Language Models (REALM 2025), pages… find
  9. 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
  10. Filippos Karolos Ventirozos, Peter Appleby, and Matthew Shardlow. 2025b. Aspect–sentiment quad prediction with distilled large language models. In Proceedings of the 15th International Conference on Recent Advances in Natural Language Processing -Natural Language Processing in the Generative AI Era,… find
  11. 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
  12. Liang-Chih Yu, Jonas Becker, Shamsuddeen Hassan Muhammad, Idris Abdulmumin, Lung-Hao Lee, Ying-Lung Lin, Jin Wang, Jan Philip Wahle, Terry Ruas, Natalia Loukachevitch, Alexander Panchenko, Ilseyar Alimova, Lilian Wanzare, Nelson Odhiambo, Bela Gipp, Kai-Wei Chang, and Saif M. Mohammad. 2026. Semeval… find
  13. Bin Liang, Hang Su, Lin Gui, Erik Cambria, and Ruifeng Xu. 2022. Aspect-based sentiment analysis via affective knowledge enhanced graph convolutional networks. Knowledge-based systems, 235:107643. find
  14. Wenxuan Zhang, Xin Li, Yang Deng, Lidong Bing, and Wai Lam. 2022. A survey on aspect-based sentiment analysis: Tasks, methods, and challenges. IEEE Transactions on Knowledge and Data Engineering, 35(11):11019–11038. find
  15. Wenxiong Liao, Bi Zeng, Xiuwen Yin, and Pengfei Wei. 2021. An improved aspect-category sentiment analysis model for text sentiment analysis based on roberta. Applied Intelligence, 51(6):3522–3533. find
  16. David Lindevelt, Suzan Verberne, and Joost Broekens. 2026. The correlation between emotion in text and speech segments is limited: A cross-modal study. In Findings of the Association for Computational Linguistics: EACL 2026, pages 2611–2621, Rabat, Morocco. Association for Computational Linguistics. find
  17. Yice Zhang, Hongling Xu, Delong Zhang, and Ruifeng Xu. 2024. A hybrid approach to dimensional aspect-based sentiment analysis using bert and large language models. Electronics, 13(18):3724. find
  18. Nimra Mughal, Ghulam Mujtaba, Sarang Shaikh, Aveenash Kumar, and Sher Muhammad Daudpota. 2024. Comparative analysis of deep natural networks and large language models for aspect-based sentiment analysis. Ieee Access, 12:60943–60959. find
  19. Ambreen Nazir, Yuan Rao, Lianwei Wu, and Ling Sun. 2020. Issues and challenges of aspect-based sentiment analysis: A comprehensive survey. IEEE Transactions on Affective Computing, 13(2):845–863. find
  20. Pablo Romero, Lifeng Han, and Goran Nenadic. 2025. Medication extraction and entity linking using Score ↓ Team 1,2408 1,2425 1,2769 1,2833 1,2942 1,3048 1,3098 1,3261 1,3283 1,3455 1,3501 1,3612 1,3654 1,3841 1,3946 1,4062 1,4109 1,4190 1,4242 1,4336 1,4394 1,4401 1,4557 1,4562 1,5073 1,5080 1,5412… find
  21. LogSigma TeleAI Bert Kittens RPI Team SokraTUM NLANGPROC ICT-NLP PICT The Classics YangS_team NTNU-SMIL PALI Habib university Cortexa SCU_Mesclab QuadAI HUS@NLP-VNU Pixel Phantoms SCUZANE UNF-BMI PAI AILS-NTUA LexiCore NCL-BU EmberAI REGLAT hdharpure DUTH surface3 2,8053 Baseline Rank 1 2 3 4 5 6 7… find

Updated

↑ back to top

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