Elamur Research⌕
☰
home / papers / 2026 / ATGL: An Adaptive-Threshold Global Loss for Document-level Relation Extraction

ATGL: An Adaptive-Threshold Global Loss for Document-level Relation Extraction

Paper · 2026 · Huangming Xu, Fu Zhang, Zhixuan Yang et al.

Summary

The paper addresses a weakness in how document-level relation extraction (DocRE) models are trained. DocRE is a multi-label task, and most current systems use the Adaptive Threshold Loss (ATL) or its variants. These losses learn a per-pair threshold class TH but optimize positive and negative relations separately. Each part interacts only with that shared threshold. The authors argue this causes two problems. First, the two parts send conflicting gradient signals that make the threshold unstable. Second, the overwhelming share of no-relation pairs pushes models toward predicting NA. Their proposed loss, ATGL, places positive, negative and threshold logits in a single shared logit space. It enforces a global ordering in which positives rank above the threshold and the threshold ranks above negatives. ATGL also adds an imbalance-aware mechanism that reduces the weight of negatives and boosts positives. The authors provide formal proofs of its optimization behavior. ATGL is model-agnostic and can be swapped into existing backbones. Across four datasets, it beats competing DocRE losses at almost no extra training cost. This makes it a practical upgrade for existing DocRE pipelines.

Key points

  • Imbalance motivation: About 94% of entity pairs in Re-DocRED carry no relation, and this skew motivates the imbalance-aware component.
  • Gains over CMM: On Re-DocRED with ATLOP (BERT base), switching from the CMM loss to ATGL raises test F1 from 76.10 to 77.26 and Ign-F1 from 74.80 to 76.05. DocuNet and KD-DocRE gain up to +0.92 F1 and +1.28 Ign-F1.
  • Comparison with LLM-based systems: TTM-RE with ATGL (RoBERTa large) reaches 82.13 F1 on Re-DocRED. This exceeds Llama3-8B-based methods such as DocKG-RAG (75.49) and EP-RSR (64.24).
  • Training cost: Training is essentially as fast as with ATL, at 40.71 vs. 40.37 minutes (ATLOP, 30 epochs).
  • Main limitation: ATGL reduces false negatives but produces more false positives than ATL, PEMSCL or HingeABL, so the precision/recall trade-off may need calibration for specific applications. Code is available at github.com/xhm-code/ATGL.

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

🎓Northeastern UniversityShenyang · China53 works · 2,205 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.Information extraction · 175 works

Cite

Cite as Huangming Xu, Fu Zhang, Zhixuan Yang, Lu Zhang and Jingwei Cheng. 2026. ATGL: An Adaptive-Threshold Global Loss for Document-level Relation Extraction. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.1603

@inproceedings{xu2026atgl,
  title = {{ATGL: An Adaptive-Threshold Global Loss for Document-level Relation Extraction}},
  author = {Huangming Xu, Fu Zhang, Zhixuan Yang, Lu Zhang and Jingwei Cheng},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1603},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1603}
}

References

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

  1. Jinheon Baek, Soyeong Jeong, Minki Kang, Jong C Park, and Sung Hwang. 2023. Knowledge-augmented language model verification. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1720–1736. find
  2. 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,…
  3. Zhichao Duan, Tengyu Pan, Zhenyu Li, Xiuxing Li, and Jianyong Wang. 2025. Comm: Concentrated margin maximization for robust document-level relation extraction. In Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), volume 39, pages 23841–23849. find
  4. Chufan Gao, Xuan Wang, and Jimeng Sun. 2024. TTM-RE: memory-augmented document-level relation extraction. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL). find
  5. Ian Goodfellow, Yoshua Bengio, and Aaron Courville. 2016. Deep learning, volume 1. find
  6. Jia Guo, Stanley Kok, and Lidong Bing. 2023. Towards integration of discriminability and robustness for document-level relation extraction. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics (EACL), pages 2598–2609. find
  7. Jiao Li, Yueping Sun, Robin J Johnson, Daniela Sci-aky, Chih-Hsuan Wei, Robert Leaman, Allan Peter Davis, Carolyn J Mattingly, Thomas C Wiegers, and Zhiyong Lu. 2016. Biocreative v cdr task corpus: a resource for chemical disease relation extraction. find
  8. Junpeng Li, Zixia Jia, and Zilong Zheng. 2023. Semi-automatic data enhancement for document-level relation extraction with distant supervision from large language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 5495–5505. find
  9. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized BERT pretraining approach. arXiv, abs/1907.11692.
  10. Youmi Ma, An Wang, and Naoaki Okazaki. 2023. Dreeam: Guiding attention with evidence for improving document-level relation extraction. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics (EACL), pages 1971–1983. find
  11. Shirui Pan, Linhao Luo, Yufei Wang, Chen Chen, Jiapu Wang, and Xindong Wu. 2024. Unifying large language models and knowledge graphs: A roadmap. IEEE Transactions on Knowledge and Data Engineering (TKDE), 36(7):3580–3599.
  12. Qingyu Tan, Ruidan He, Lidong Bing, and Hwee Tou Ng. 2022a. Document-level relation extraction with adaptive focal loss and knowledge distillation. In Findings of the Annual Meeting of the Association for Computational Linguistics (ACL), pages 1672– 1681.
  13. Qingyu Tan, Lu Xu, Lidong Bing, Hwee Tou Ng, and Sharifah Mahani Aljunied. 2022b. Revisiting docred-addressing the false negative problem in relation extraction. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 8472–8487.
  14. Khai Phan Tran, Wen Hua, and Xue Li. 2025. Vaediff-docre: End-to-end data augmentation framework for document-level relation extraction. In Proceedings of the 31st International Conference on Computational Linguistics (COLING), pages 7307–7320. find
  15. Jize Wang, Xinyi Le, Xiaodi Peng, and Cailian Chen. 2023. Adaptive hinge balance loss for document-level relation extraction. In Findings of Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 3872–3878. find
  16. Ye Wang, Xinxin Liu, Wenxin Hu, and Tao Zhang. 2022. A unified positive-unlabeled learning framework for document-level relation extraction with different levels of labeling. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 4123–4135. find
  17. Ying Wei and Qi Li. 2022. Sagdre: Sequence-aware graph-based document-level relation extraction with adaptive margin loss. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), pages 2000–2008. find
  18. Huangming Xu, Fu Zhang, and Jingwei Cheng. 2025a. An adaptive multi-threshold loss and a general framework for collaborating losses in document-level relation extraction. In Findings of the Association for Computational Linguistics (ACL), pages 20996– 21007. find
  19. Huangming Xu, Fu Zhang, Jingwei Cheng, and Xin Li. 2025b. ARPDL: adaptive relational prior distribution loss as an adapter for document-level relation extraction. In Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI), pages 8313–8321. find
  20. Xiaolong Xu, Chenbin Li, Haolong Xiang, Lianyong Qi, Xuyun Zhang, and Wanchun Dou. 2024. Attention based document-level relation extraction with none class ranking loss. In Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence (IJCAI), pages 6569–6577. find
  21. Xiaolong Xu, Yibo Zhou, Haolong Xiang, Xiaoyong Li, Xuyun Zhang, Lianyong Qi, and Wanchun Dou. 2025c. Docks-rag: Optimizing document-level relation extraction through llm-enhanced hybrid prompt tuning. In Forty-second International Conference on Machine Learning, ICML 2025, Vancouver, BC, Canada, Ju… find
  22. Yuan Yao, Deming Ye, Peng Li, Xu Han, Yankai Lin, Zhenghao Liu, Zhiyuan Liu, Lixin Huang, Jie Zhou, and Maosong Sun. 2019. Docred: A large-scale document-level relation extraction dataset. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL), pages 764–777… find
  23. Klim Zaporojets, Johannes Deleu, Chris Develder, and Thomas Demeester. 2021. DWIE: an entity-centric dataset for multi-task document-level information extraction. Inf. Process. Manag., 58(4):102563. find
  24. Shuang Zeng, Runxin Xu, Baobao Chang, and Lei Li. 2020. Double graph based reasoning for document-level relation extraction. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1630–1640.
  25. Bowen Zhang and Harold Soh. 2024. Extract, define, canonicalize: An llm-based framework for knowledge graph construction. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9820–9836.
  26. Fu Zhang, Qi Miao, Jingwei Cheng, Hongsen Yu, Yi Yan, Xin Li, and Yongxue Wu. 2024. Srf: enhancing document-level relation extraction with a novel secondary reasoning framework. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 15426–15439. find
  27. Fu Zhang, Hongsen Yu, Jingwei Cheng, and Huangming Xu. 2025. Entity pair-guided relation summarization and retrieval in llms for document-level relation extraction. In Findings of the Association for Computational Linguistics: NAACL 2025, pages 4022–4037. find
  28. Liang Zhang, Zijun Min, Jinsong Su, Pei Yu, Ante Wang, and Yidong Chen. 2023. Exploring effective inter-encoder semantic interaction for document-level relation extraction. In Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence (IJCAI), pages 5278–5286. find
  29. Ningyu Zhang, Xiang Chen, Xin Xie, Shumin Deng, Chuanqi Tan, Mosha Chen, Fei Huang, Luo Si, and Huajun Chen. 2021. Document-level relation extraction as semantic segmentation. In Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence (IJCAI), pages 3999–4006. find
  30. Wenxuan Zhou, Kevin Huang, Tengyu Ma, and Jing Huang. 2021. Document-level relation extraction with adaptive thresholding and localized context pooling. In Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), volume 35, pages 14612– 14620. find
  31. Yang Zhou and Wee Sun Lee. 2022. None class ranking loss for document-level relation extraction. In Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence (IJCAI), pages 4538–4544. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xu-2026-atgl-adaptive-threshold-global-loss/