Elamur Research⌕
☰
home / papers / 2026 / Team HITS at SemEval-2026 Task 4:Enhancing narrative text embedding model training with hard negatives generation and self-distillation

Team HITS at SemEval-2026 Task 4:Enhancing narrative text embedding model training with hard negatives generation and self-distillation

Paper · 2026 · Qian Zhou, Yi Fan, Wei Liu et al.

Summary

This paper describes HITS's system for Track B of SemEval-2026 Task 4. The track asks for story embeddings whose similarity reflects narrative properties: theme, course of action and outcome, rather than surface wording. The authors identify two problems with standard contrastive fine-tuning. Random negatives are too easy, and narrative similarity has several dimensions. To address this, they prompt Qwen2.5-32B-Instruct to write synthetic hard negatives for each anchor. Each negative keeps exactly one dimension (theme, structure or outcome) and alters the other two.

They then fine-tune Qwen3-Embedding-8B with LoRA, using an InfoNCE loss that combines these generated negatives with in-batch negatives. The loss is supplemented by KL-divergence self-distillation from a teacher with the same architecture. The best-performing model is promoted to teacher in each round, and a teacher-based mask removes likely false negatives. Training uses the task's 1,900 LLM-generated triples, and evaluation measures triplet accuracy. The submitted single-iteration distillation model reached 0.705 test accuracy and ranked third in the track. The results suggest that dimension-controlled negatives and soft labels improve narrative representations.

Key points

  • Training data comprised a 39-triple sample, a 200-triple development set and 1,900 synthetic triples. The Track B test set contains 849 individual stories.
  • Hard negatives were generated with Qwen2.5-32B-Instruct (temperature 0.8, top_p 0.9), producing three candidates per controlled narrative dimension.
  • The backbone was Qwen3-Embedding-8B with LoRA (rank 32, α=64, dropout 0.1) applied to the attention and FFN projections. Training used AdamW at a learning rate of 5e-5 for 5 epochs, batch size 8 and a maximum length of 1024. The false-negative mask used a margin of −0.05.
  • Test accuracy rose from 0.6675 for the baseline to 0.6750 with multi-negative training, then to 0.7050 with one round of self-distillation. A second distillation iteration scored highest on validation (0.74) but fell to 0.69 on test, which the authors attribute to overfitting and to bias from the synthetic data.
  • Code is released at github.com/zq03-web/sts_embedding. The stated limitations are that the method was evaluated on only one test set and that larger generator models were not tried.

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

🎓Heidelberg UniversityHeidelberg · Germany39 works · 41,356 citations🔬Heidelberg Institute for Theoretical StudiesHeidelberg · Germany31 works · 1,845 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

Cite

Cite as Qian Zhou, Yi Fan, Wei Liu and Michael Strube. 2026. Team HITS at SemEval-2026 Task 4:Enhancing narrative text embedding model training with hard negatives generation and self-distillation. 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-1.338

@inproceedings{zhou2026team,
  title = {{Team HITS at SemEval-2026 Task 4:Enhancing narrative text embedding model training with hard negatives generation and self-distillation}},
  author = {Qian Zhou, Yi Fan, Wei Liu and Michael Strube},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.338},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.338}
}

References

17 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. 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,…
  2. Tommaso Furlanello, Zachary Lipton, Michael Tschannen, Laurent Itti, and Anima Anandkumar. 2018. Born again neural networks. In Proceedings of the 35th International Conference on Machine Learning, volume 80 of Proceedings of Machine Learning Research, pages 1607–1616. PMLR. find
  3. Tianyu Gao, Xingcheng Yao, and Danqi Chen. 2021. SimCSE: Simple contrastive learning of sentence embeddings. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 6894–6910, Online and Punta Cana, Dominican Republic. Association for Computational Linguistic…
  4. Hans Ole Hatzel, Ekaterina Artemova, Haimo Stiemer, Evelyn Gius, and Chris Biemann. 2026. SemEval-2025 task 4: Narrative similarity and narrative representation learning. In Proceedings of the 20th International Workshop on Semantic Evaluation (SemEval-2026), San Diego, CA, USA. find
  5. Hans Ole Hatzel and Chris Biemann. 2024. Story embeddings — narrative-focused representations of fictional stories. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 5931–5943, Miami, Florida, USA. Association for Computational Linguistics. find
  6. Geoffrey Hinton, Oriol Vinyals, and Jeffrey Dean. 2015. Distilling the knowledge in a neural network. In NIPS Deep Learning and Representation Learning Workshop.
  7. Edward J Hu, yelong shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. LoRA: Low-rank adaptation of large language models. In International Conference on Learning Representations.
  8. Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. Dense passage retrieval for open-domain question answering. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 6769–6781, Onli…
  9. Solomon Kullback and Richard A Leibler. 1951. On information and sufficiency. The annals of mathematical statistics, 22(1):79–86. find
  10. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. In Advances in Neural Information Proc…
  11. Xiaopeng Li, Xiangyang Li, Hao Zhang, Zhaocheng Du, Pengyue Jia, Yichao Wang, Xiangyu Zhao, Huifeng Guo, and Ruiming Tang. 2024. Syneg: Llm-driven synthetic hard-negatives for dense retrieval. arXiv preprint arXiv:2412.17250. — [arXiv:2412.17250](https://arxiv.org/abs/2412.17250) find
  12. Niklas Muennighoff, Nouamane Tazi, Loic Magne, and Nils Reimers. 2023. MTEB: Massive text embedding benchmark. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 2014–2037, Dubrovnik, Croatia. Association for Computational Linguistic…
  13. Qwen, :, 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, and 25 others. 2025. Qwen2.5 technical report. Preprint, arXiv:2412.15115. find
  14. Florian Schroff, Dmitry Kalenichenko, and James Philbin. 2015. Facenet: A unified embedding for face recognition and clustering. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 815–823.
  15. Aaron van den Oord, Yazhe Li, and Oriol Vinyals. 2019. Representation learning with contrastive predictive coding. Preprint, arXiv:1807.03748.
  16. Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, and Furu Wei. 2024. Improving text embeddings with large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 11897–11916, Bangkok, Thailand. Asso…
  17. Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou. 2025. Qwen3 embedding: Advancing text embedding and reranking through foundation models. Preprint, arXiv:2506.05176. *Figure 2: NEG_THEME Prompt*

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhou-2026-team-hits-semeval-2026-task/