Elamur Research⌕
☰
home / papers / 2026 / Team BOBW (Best Of Both Worlds) at SemEval-2026 Task 3: Modular Cross-Attention Encoders for Dimensional Aspect-Based Sentiment Analysis

Team BOBW (Best Of Both Worlds) at SemEval-2026 Task 3: Modular Cross-Attention Encoders for Dimensional Aspect-Based Sentiment Analysis

Paper · 2026 · Michal Rynowiecki and Rob van der Goot

Summary

The paper describes an encoder-only system for SemEval-2026 Task 3 (DiMBaSA). The task asks systems to extract quadruplets from English laptop and restaurant reviews. Each quadruplet holds an aspect, an opinion, a category, and valence–arousal scores on a 1–9 scale. The authors chain four stages:

  • BIO+CRF span tagging of aspects and opinions,
  • binary validation of candidate aspect–opinion pairs,
  • category classification,
  • valence–arousal regression.

Most stages use two BERT encoders, where one encoder queries the other's hidden states through inserted cross-attention layers. This setup was compared with a baseline that adds self-attention layers with matched parameter counts.

Cross-attention gave only marginal gains, and on laptops the self-attention variant was slightly better. Errors carried forward through the pipeline, mostly from the pair-validation step and from fine-grained laptop categories. Results changed little with model size: BERT-medium performed about as well as BERT-large-cased. This suggests that small modular encoders can serve as a cheap alternative to large decoder LLMs for structured sentiment extraction.

Key points

  • Best cF1 scores: 0.53 on restaurant data (BERT-Medium-CA, the official submission) and 0.26 on laptop data (BERT-Large-Cased).
  • LLM baselines: Kimi K2 scored 0.37 on restaurant and 0.28 on laptop, so it slightly beat the system on laptops. Qwen-3 14B scored 0.28 and 0.15.
  • Cross-attention placement: layers were inserted after layers 3 and 7 for BERT-medium and arctic-embed, and after layers 8, 14 and 23 for BERT-large-cased. They did not noticeably change training time.
  • Ablation: replacing BERT-tiny with DeBERTa-v3 Large raised aspect/opinion precision from 0.65 to 0.84. Overall results still barely improved because of error propagation.
  • Category accuracy: the first category component reached about 0.80 accuracy, while the second laptop subcategory fell to about 0.50. Valence–arousal prediction was reliable, including on sarcastic reviews. Snowflake Arctic Embed performed poorly (cF1 0.15 on laptop).

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

🎓IT University of CopenhagenCopenhagen · Denmark27 works · 2,406 citations

 Fields, topics and concepts

📚 Shelves 5The 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 worksComputer Vision and Pattern Recognition · 3,486 works

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

Cite

Cite as Michal Rynowiecki and Rob van der Goot. 2026. Team BOBW (Best Of Both Worlds) at SemEval-2026 Task 3: Modular Cross-Attention Encoders 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-1.179

@inproceedings{rynowiecki2026team,
  title = {{Team BOBW (Best Of Both Worlds) at SemEval-2026 Task 3: Modular Cross-Attention Encoders for Dimensional Aspect-Based Sentiment Analysis}},
  author = {Michal Rynowiecki and Rob {van der Goot}},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.179},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.179}
}

References

19 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jeremy Barnes, Laura Oberlaender, Enrica Troiano, Andrey Kutuzov, Jan Buchmann, Rodrigo Agerri, Lilja Øvrelid, and Erik Velldal. 2022. find
  2. SemEval 2022 task 10: Structured sentiment analysis. In Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 1280–1295, Seattle, United States. Association for Computational Linguistics. find
  3. 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
  4. Prajjwal Bhargava, Aleksandr Drozd, and Anna Rogers. 2021. Generalization in nli: Ways (not) to go beyond simple heuristics. Preprint, arXiv:2110.01518.
  5. Rich Caruana. 1997. Multitask learning. Mach. Learn., 28(1):41–75. find
  6. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. Cite arxiv:1810.04805.
  7. Minqing Hu and Bing Liu. 2004. Mining and summarizing customer reviews. In Proceedings of the Tenth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’04, page 168–177, New York, NY, USA. Association for Computing Machinery. find
  8. John D. Lafferty, Andrew McCallum, and Fernando C. N. Pereira. 2001. Conditional random fields: Probabilistic models for segmenting and labeling sequence data. In Proceedings of the Eighteenth International Conference on Machine Learning, ICML ’01, page 282–289, San Francisco, CA, USA. Morgan Kaufma… find
  9. Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini. 2009. The graph neural network model. Trans. Neur. Netw., 20(1):61–80. find
  10. Iulia Turc, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Well-read students learn better: The impact of student initialization on knowledge distillation. CoRR, abs/1908.08962.
  11. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in neural information processing systems, pages 5998–6008.
  12. 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
  13. 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
  14. Luke Merrick, Danmei Xu, Gaurav Nuti, and Daniel Campos. 2024. Arctic-embed: Scalable, efficient, and accurate text embedding models. Preprint, arXiv:2405.05374. — [arXiv:2405.05374](https://arxiv.org/abs/2405.05374) find
  15. Bo Pang, Lillian Lee, and Shivakumar Vaithyanathan. 2002. Thumbs up? sentiment classification using machine learning techniques. In Proceedings of the 2002 Conference on Empirical Methods in Natural Language Processing (EMNLP 2002), pages 79–86. Association for Computational Linguistics.
  16. 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…
  17. Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. 2018. Improving language understanding by generative pre-training. find
  18. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. Preprint, arXiv:1908.10084.
  19. James A. Russell. 1980. A circumplex model of affect. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/rynowiecki-2026-team-bobw-best-both-worlds/