Elamur Research⌕
☰
home / papers / 2026 / Entity-Level Sentiment Analysis with Sentence Relevance Detection

Entity-Level Sentiment Analysis with Sentence Relevance Detection

Summary

This paper addresses entity-level sentiment analysis (Elsa), where a model must assign a document-level polarity to each person or organization mentioned in a text. Most Norwegian Elsa reviews (drawn from NoReC) exceed the 512-token limit of encoders such as NorBert3-large. Rather than chunking documents and aggregating predictions, the authors train a model to keep only the sentences or spans relevant to a given entity. This compressed input lets encoders read the whole relevant context, and it also shortens training and inference for decoders.

For sub-10B decoder models, the paper compares prompting with two fine-tuning regimes, supervised fine-tuning (SFT) and pairwise comparison (PC). PC is borrowed from reward modeling: the model learns to prefer a text stating the correct label over one stating a wrong label.

Both fine-tuned decoders beat earlier Elsa baselines and fine-tuned encoders. PC is simpler, faster and more stable than SFT, and it also performs strongly on PerSenT and on EuroEval sentiment tasks in Norwegian and English. This makes PC a practical new approach to text classification.

Key points

  • Datasets: The Norwegian Elsa dataset has 2,479 sentiment-annotated entities, and about 60% of its training texts exceed 512 tokens. The English PerSenT dataset has about 25% of texts over that limit. Earlier Elsa baselines reached 68.1% weighted F1 (NorBert3-large, sentence aggregation) and 73.3% (zero-shot GPT-4).
  • PC setup: With gemma-2-9b, building two training pairs per entity (one for each wrong label) raised weighted F1 from 80.15% to 81.72%. Two training epochs worked best, with eval accuracy of 0.875.
  • Prompt language and efficiency: Norwegian prompts averaged 83.44% accuracy versus 82.20% for English prompts. The authors fine-tuned with LoRA (rank 16, alpha 32). 4-bit quantization cost little, giving 82.25% versus 82.87% at 16-bit precision.
  • Compression check: In a manual review of 50 non-neutral entities, the sentiment signal stayed intact in 43 cases while texts shrank to 40–64% of their original length. The ent2ent heuristic lost the signal twice, both times when the sentiment appeared before the entity was first named.
  • Code: Available at github.com/egilron/elsa-lrec.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/35ideqx4jk89

@inproceedings{ronningstad2026entity,
  title = {{Entity-Level Sentiment Analysis with Sentence Relevance Detection}},
  author = {Rønningstad, Egil and Klinger, Roman and Øvrelid, Lilja and Velldal, Erik},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/35ideqx4jk89},
  url = {https://doi.org/10.63317/35ideqx4jk89}
}

References

55 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jason Ansel, Edward Yang, Horace He, Natalia Gimelshein, Animesh Jain, Michael Voznesensky, Bin Bao, Peter Bell, David Berard, Evgeni Burovski, et al. 2024. Pytorch 2: Faster machine learning through dynamic python bytecode transformation and graph compilation. In Proceedings of the 29th ACM Interna… find
  2. Jeremy Barnes, Laura Oberlaender, Enrica Troiano, Andrey Kutuzov, Jan Buchmann, Rodrigo Agerri, Lilja Øvrelid, and Erik Velldal. 2022. find
  3. 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
  4. Mohaddeseh Bastan, Mahnaz Koupaee, Youngseo Son, Richard Sicoli, and Niranjan Balasubramanian. 2020. Author’s sentiment prediction. In Proceedings of the 28th International Conference on Computational Linguistics, pages 604–615, Barcelona, Spain (Online). International Committee on Computational Lin… find
  5. Iz Beltagy, Matthew E Peters, and Arman Cohan. 2020. Longformer: The long-document transformer. arXiv preprint arXiv:2004.05150. find
  6. Zvi Ben-Ami, Ronen Feldman, and Benjamin Rosenfeld. 2015. Exploiting the focus of the document for enhanced entities’ sentiment relevance detection. In 2015 IEEE International Conference on Data Mining Workshop (ICDMW), pages 1284–1293. find
  7. Ralph Allan Bradley and Milton E Terry. 1952. Rank analysis of incomplete block designs: I. the method of paired comparisons. Biometrika, 39(3/4):324–345. find
  8. Martin Juan José Bucher and Marco Martini. 2024. Fine-tuned’small’llms (still) significantly outperform zero-shot generative ai models in text classification. arXiv preprint arXiv:2406.08660. — [arXiv:2406.08660](https://arxiv.org/abs/2406.08660) find
  9. Hongjie Cai, Heqing Ma, Jianfei Yu, and Rui Xia. 2024. A joint coreference-aware approach to document-level target sentiment analysis. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 12149–12160, Bangkok, Thailand. Association… find
  10. Lingjia Deng and Janyce Wiebe. 2015. MPQA 3.0: An entity/event-level sentiment corpus. In Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1323–1328, Denver, Colorado. Association for Computational L… find
  11. Franck Dernoncourt, Mohammad Ghassemi, and Walter Chang. 2018. A repository of corpora for summarization. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA). find
  12. James C. Douglas, Yidong Gan, Ben Hachey, and Jonathan K. Kummerfeld. 2025. Less is more: Explainable and efficient ICD code prediction with clinical entities. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 30835–30847, Vienn… find
  13. Joe Ellis, Jeremy Getman, and Stephanie M Strassel. 2014. Overview of linguistic resources for the tac kbp 2014 evaluations: Planning, execution, and results. In Proceedings of TAC KBP 2014 Workshop, National Institute of Standards and Technology, pages 17–18. find
  14. Francisco Guzmán, Shafiq Joty, Lluís Màrquez, and Preslav Nakov. 2015. Pairwise neural machine translation evaluation. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long… find
  15. Charles R Harris, K Jarrod Millman, Stéfan J Van Der Walt, Ralf Gommers, Pauli Virtanen, David Cournapeau, Eric Wieser, Julian Taylor, Sebastian Berg, Nathaniel J Smith, et al. 2020. Array programming with numpy. nature, 585(7825):357–362. find our card
  16. Mingqian He, Fei Zhao, Chonggang Lu, Ziyan Liu, Yue Wang, and Haofu Qian. 2025. Gencls++: Pushing the boundaries of generative classification in llms through comprehensive sft and rl studies across diverse datasets. arXiv preprint arXiv:2504.19898. — [arXiv:2504.19898](https://arxiv.org/abs/2504.19898) find
  17. Pengcheng He, Jianfeng Gao, and Weizhu Chen. 2021. Debertav3: Improving deberta using electra-style pre-training with gradient-disentangled embedding sharing. find our card
  18. 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, pages 168–177. find
  19. J. D. Hunter. 2007. Matplotlib: A 2d graphics environment. Computing in Science & Engineering, 9(3):90–95. find
  20. Jan Kocoń, Igor Cichecki, Oliwier Kaszyca, Mateusz Kochanek, Dominika Szydło, Joanna Baran, Julita Bielaniewicz, Marcin Gruza, Arkadiusz Janz, Kamil Kanclerz, et al. 2023. Chatgpt: Jack of all trades, master of none. Information Fusion, 99:101861. find our card
  21. Alapan Kuila and Sudeshna Sarkar. 2024a. Deciphering political entity sentiment in news with large language models: Zero-shot and few-shot strategies. In Proceedings of the Second Workshop on Natural Language Processing for Political Sciences @ LREC-COLING 2024, pages 1–11, Torino, Italia. ELRA and… find
  22. Alapan Kuila and Sudeshna Sarkar. 2024b. Deciphering political entity sentiment in news with large language models: Zero-shot and few-shot strategies. arXiv preprint arXiv:2404.04361. find
  23. Andrey Kutuzov, Jeremy Barnes, Erik Velldal, Lilja Øvrelid, and Stephan Oepen. 2021. Large-scale contextualised language modelling for norwegian. In Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa 2021). find
  24. Jan Leike, David Krueger, Tom Everitt, Miljan Martic, Vishal Maini, and Shane Legg. 2018. Scalable agent alignment via reward modeling: a research direction. arXiv preprint arXiv:1811.07871. find
  25. Yun Luo, Hongjie Cai, Linyi Yang, Yanxia Qin, Rui Xia, and Yue Zhang. 2022. Challenges for open-domain targeted sentiment analysis. find
  26. Ibraheem Muhammad Moosa, Rui Zhang, and Wenpeng Yin. 2024. Mt-ranker: Reference-free machine translation evaluation by inter-system ranking. arXiv preprint arXiv:2401.17099. find
  27. Michael Oliver and Guan Wang. 2024. Crafting efficient fine-tuning strategies for large language models. arXiv preprint arXiv:2407.13906. — [arXiv:2407.13906](https://arxiv.org/abs/2407.13906) find
  28. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F Christiano, Jan Leike, and Ryan Lowe. 2022a. Tr… find
  29. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. 2022b. Training language models to follow instructions with human feedback. Advances in neural information processing systems, 35:27730–27744. find
  30. Lilja Øvrelid, Petter Mæhlum, Jeremy Barnes, and Erik Velldal. 2020. A fine-grained sentiment dataset for Norwegian. In Proceedings of the 12th Edition of the Language Resources and Evaluation Conference, Marseille, France, 2020. find
  31. The pandas development team. 2020. pandas-dev/pandas: Pandas. find
  32. Aldo Pareja, Nikhil Shivakumar Nayak, Hao Wang, Krishnateja Killamsetty, Shivchander Sudalairaj, Wenlong Zhao, Seungwook Han, Abhishek Bhandwaldar, Guangxuan Xu, Kai Xu, et al. 2024. Unveiling the secret recipe: A guide for supervised fine-tuning small llms. arXiv preprint arXiv:2412.13337. — [arXiv:2412.13337](https://arxiv.org/abs/2412.13337) find
  33. Hyunji Park, Yogarshi Vyas, and Kashif Shah. 2022. Efficient classification of long documents using transformers. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 702–709, Dublin, Ireland. Association for Computational Linguis… find
  34. F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. 2011. Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12:282… find
  35. Qiankun Pi, Jicang Lu, Taojie Zhu, Yepeng Sun, Shunhang Li, and Jiaxing Guo. 2024. Enhancing cross-evidence reasoning graph for document-level relation extraction. PeerJ Computer Science, 10:e2123. find
  36. 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… find our card
  37. Egil Rønningstad, Roman Klinger, Lilja Øvrelid, and Erik Velldal. 2024. Entity-level sentiment: More than the sum of its parts. In Proceedings of the 14th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis, pages 84–96, Bangkok, Thailand. Association for Computa… find
  38. Egil Rønningstad and Gaurav Negi. 2025. Ltg at semeval-2025 task 10: Optimizing context for classification of narrative roles. find
  39. Dan Saattrup Nielsen, Kenneth Enevoldsen, and Peter Schneider-Kamp. 2025. Encoder vs decoder: Comparative analysis of encoder and decoder language models on multilingual NLU tasks. In Proceedings of the Joint 25th Nordic Conference on Computational Linguistics and 11th Baltic Conference on Human Lan… find
  40. David Samuel, Andrey Kutuzov, Samia Touileb, Erik Velldal, Lilja Øvrelid, Egil Rønningstad, Elina Sigdel, and Anna Palatkina. 2023. NorBench – a benchmark for Norwegian language models. In Proceedings of the 24th Nordic Conference on Computational Linguistics (NoDaLiDa), pages 618–633, Tórshavn, Far… find
  41. Boheng Sheng, Jiacheng Yao, Meicong Zhang, and Guoxiu He. 2025. Dynamic chunking and selection for reading comprehension of ultra-long context in large language models. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 31857–318… find
  42. Richard Socher, Alex Perelygin, Jean Wu, Jason Chuang, Christopher D. Manning, Andrew Ng, and Christopher Potts. 2013. Recursive deep models for semantic compositionality over a sentiment treebank. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1631–… find
  43. Yixuan Tang, Yi Yang, Allen Huang, Andy Tam, and Justin Tang. 2023. FinEntity: Entity-level sentiment classification for financial texts. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 15465–15471, Singapore. Association for Computational Linguistics… find
  44. Erik Velldal, Lilja Øvrelid, Eivind Alexander Bergem, Cathrine Stadsnes, Samia Touileb, and Fredrik Jørgensen. 2018. NoReC: The Norwegian review corpus. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resour… find
  45. Leandro von Werra, Younes Belkada, Lewis Tunstall, Edward Beeching, Tristan Thrush, and Nathan Lambert. 2022. Trl: Transformer reinforcement learning. https://github.com/huggingface/trl. ↗
  46. Binghai Wang, Rui Zheng, Lu Chen, Zhiheng Xi, Wei Shen, Yuhao Zhou, Dong Yan, Tao Gui, Qi Zhang, and Xuanjing Huang. 2024. Reward modeling requires automatic adjustment based on data quality. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 4041–4064, Miami, Florida, U… find
  47. Xiang Wei, Xingyu Cui, Ning Cheng, Xiaobin Wang, Xin Zhang, Shen Huang, Pengjun Xie, Jinan Xu, Yufeng Chen, Meishan Zhang, et al. 2023. Chatie: Zero-shot information extraction via chatting with chatgpt. arXiv preprint arXiv:2302.10205. find our card
  48. Wes McKinney. 2010. Data Structures for Statistical Computing in Python. In Proceedings of the 9th Python in Science Conference, pages 56 – 61. find
  49. Janyce Wiebe, Theresa Wilson, and Claire Cardie. 2005. Annotating expressions of opinions and emotions in language. Language resources and evaluation, 39(2):165–210. find
  50. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, et al. 2019. Huggingface’s transformers: State-of-the-art natural language processing. arXiv preprint arXiv:1910.03771. find our card
  51. Junting Ye and Steven Skiena. 2019. Mediarank: Computational ranking of online news sources. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2469–2477. 9. find
  52. Language Resource References Bastan, Mohaddeseh and Koupaee, Mahnaz and Son, Youngseo and Sicoli, Richard and Balasubramanian, Niranjan. 2020. PerSenT. Stony Brook University. https://stonybrooknlp.github.io/PerSenT/. ↗
  53. Lilja Øvrelid and Petter Mæhlum and Jeremy Barnes and Erik Velldal. 2020. NoReC sentence. Language Technology Group, University of Oslo. https://huggingface.co/datasets/ltg/norec_sentence. ↗
  54. Rønningstad, Egil and Øvrelid, Lilja and Velldal, Erik. 2024. ELSA: An Entity-Level Sentiment Analysis Dataset for Norwegian. Language Technology Group, University of Oslo. https://github.com/ltgoslo/ELSA. ↗
  55. Richard Socher and Alex Perelygin and Jean Wu and Jason Chuang and Christopher Manning and Andrew Ng and Christopher Potts. 2013. Stanford Sentiment Treebank. The Stanford Natural Language Processing Group. https://nlp.stanford.edu/sentiment/. ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ronningstad-2026-entity-level-sentiment-analysis-sentence/