Elamur Research⌕
☰
home / papers / 2026 / TeamLasse at SemEval-2026 Task 3: A Hybrid Generative-Discriminative Framework for Dimensional Aspect-Based Sentiment Analysis

TeamLasse at SemEval-2026 Task 3: A Hybrid Generative-Discriminative Framework for Dimensional Aspect-Based Sentiment Analysis

Paper · 2026 · Lasse Strothe

Summary

This paper presents the author's submission to Track A of SemEval-2026 Task 3, which covers Dimensional Aspect-Based Sentiment Analysis (DimABSA). The task pairs sentiment elements (aspects, opinions, categories) with continuous Valence-Arousal (VA) scores across six languages, including Ukrainian and Tatar, and several domains. Instead of building a separate model per subtask, the author uses one two-stage pipeline. First, a fine-tuned Qwen2.5-32B-Instruct generates the structured tuples. Second, an XLM-RoBERTa-large cross-encoder regresses VA from the sentence plus the extracted aspect and opinion. The rationale is that autoregressive LLMs handle numbers as tokens and are poorly suited to continuous scoring. Training used only the Subtask 3 data, with upsampling to balance domains and low-resource languages. The system performs well on extraction, ranking 2nd in Japanese Hotel. On Subtask 1, where only aspects are given, it is weaker: without an explicit opinion term, Arousal prediction degrades sharply. The work illustrates both the strengths and the train–inference mismatch risks of decoupled pipelines for multilingual ABSA.

Key points

  • Data: The official DimABSA dataset covers English, Chinese, Japanese, Russian, Tatar and Ukrainian. Training relied solely on the Subtask 3 (DimASQP) train and dev sets, with no external data. Finance was handled zero-shot.
  • Balancing: Upsampling worked at two levels:
    • Hospitality vs. Technology meta-domains, balanced within each language.
    • Languages, with smaller ones raised to the size of the largest.
  • Training setup:
    • Qwen2.5-32B was tuned with 4-bit NF4, LoRA (r=16, α=16) and Unsloth on one A100 40GB for one epoch.
    • XLM-R was trained for 4 epochs with MSE loss, and outputs were clamped to [1, 9].
  • Subtasks 2 and 3 (cF1):
    • Japanese Hotel scored 0.5694 on Subtask 2, ranking 2nd.
    • Ukrainian Restaurant scored 0.5270 on Subtask 2 (8th) and 0.4879 on Subtask 3 (5th).
    • English Laptop fell from 0.5513 to 0.3049, attributed to overlapping categories.
  • Subtask 1: Ranks were 12th–27th, and Valence correlation stayed far above Arousal correlation.
    • Ukrainian Restaurant: PCC_V 0.8059, PCC_A 0.4745.
    • Japanese Finance: PCC_V 0.7937, PCC_A 0.2114.

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

🎓Technical University of MunichMunich · Germany62 works · 48,970 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 Lasse Strothe. 2026. TeamLasse at SemEval-2026 Task 3: A Hybrid Generative-Discriminative Framework 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-1.273

@inproceedings{strothe2026teamlasse,
  title = {{TeamLasse at SemEval-2026 Task 3: A Hybrid Generative-Discriminative Framework for Dimensional Aspect-Based Sentiment Analysis}},
  author = {Lasse Strothe},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.273},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.273}
}

References

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

  1. J. A. Russell. 1980. A circumplex model of affect. Journal of Personality and Social Psychology, 39(6):1161– 1178. find
  2. 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…
  3. 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…
  4. Cem Üyük, Danica Rovó, Shaghayeghkolli, Rabia Varol, Georg Groh, and Daryna Dementieva. 2024. Crafting tomorrow’s headlines: Neural news generation and detection in English, Turkish, Hungarian, and Persian. In Proceedings of the Third Workshop on NLP for Positive Impact, pages 271–307, Miami, Florid… find
  5. Daniel Han and Michael Han. 2023. Unsloth: Accelerating large language model fine-tuning. https://github.com/unslothai/unsloth. ↗
  6. An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, and 23 others. 2025. Qwen2.5 technical report. Preprint, arXiv:241… find
  7. Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Liang Wang, Weizhu Chen, and 1 others. 2022. Lora: Low-rank adaptation of large language models. Iclr, 1(2):3.
  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. Shaghayegh Kolli, Richard Rosenbaum, Timo Cavelius, Lasse Strothe, Andrii Lata, and Jana Diesner. 2025. Hybrid fact-checking that integrates knowledge graphs, large language models, and search-based retrieval agents improves interpretable claim verification. In Proceedings of the 9th Widening NLP Wo… find
  10. 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
  11. 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, Online and Punta Cana, Dominican Republic. Association…
  12. 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), find

Updated

↑ back to top

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