Elamur Research⌕
☰
home / papers / 2026 / Enhanced Reasoning for Biomedical Document-Level Relation Extraction via a Novel Cascade Language Model Framework

Enhanced Reasoning for Biomedical Document-Level Relation Extraction via a Novel Cascade Language Model Framework

Paper · 2026 · Haohua Song, Wenhao Gu, Zhijing Li et al.

Summary

Extracting relations between biomedical entities across a whole document is harder than doing it within one sentence. It requires linking evidence spread over several sentences, resolving synonymous mentions such as phenylketonuria and PKU, and applying specialised knowledge. Fine-tuned PLMs are cheap and accurate on easy cases but reason poorly. LLMs reason better, yet in this domain they tend to hallucinate relations that do not exist. The authors propose CoRE, a cascade built on a detect-then-rethink idea. A domain-specific PLM first scores every entity pair and keeps the predictions it is confident about. Uncertain pairs are passed to an LLM, which receives few-shot demonstrations chosen by a document-level semantic retriever. The LLM then reasons through an iterative chain-of-thought (ItCoT) procedure that consults a dynamic library of past errors to reflect on and correct its answers. On BioRED and CDR, CoRE beats PLM-only, LLM-only and naive cascade baselines while calling the LLM far less often. This makes LLM-assisted literature mining more accurate and cheaper to deploy.

Key points

  • Routing signal: confidence is the PLM's highest relation probability. On the CDR dev set, DocuNet's high-confidence predictions proved much more reliable than its low-confidence ones. LLMs, by contrast, over-predicted relations on no-relation pairs.
  • Overall F1 against a cascaded GPT-RE baseline (Table 5):
    • CDR: CoRE 88.20 vs. 86.03 for Cascade + GPT-RE.
    • BioRED: CoRE 66.35 vs. 62.79 for Cascade + GPT-RE.
    • On the low-confidence subsets, CoRE's average F1 gain over standalone GPT-RE was +5.39 on CDR and +5.30 on BioRED.
  • Hallucination analysis (CDR low-confidence subset):
    • With Deepseek-v3.2, the false positive rate (FPR) fell from 41.42% to 26.19% and the false discovery rate (FDR) from 48.45% to 37.28%.
    • With GPT-4o, FPR fell from 25.84% to 22.83%.
    • Naive CoT actually increased false positives compared with the plain baseline.
  • Efficiency on the CDR test set: selective routing cut API cost by about 45.55% ($54.7 → $29.8) and latency by about 46.89% (1,220 → 648 minutes).
  • Limitations:
    • The routing threshold must be tuned by hand on a validation set.
    • Retrieval depends on the training set containing relevant examples.
    • Low-confidence samples incur extra latency.

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

🎓Shenzhen UniversityShenzhen · China16 works · 95 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 worksBiochemistry, Genetics and Molecular Biology · 234 worksMolecular Biology · 231 works

🧭 Topics 3Research subjects and languages the work is about.Language models & pretraining · 3,320 worksReasoning · 1,426 worksInformation extraction · 167 works

Cite

Cite as Haohua Song, Wenhao Gu, Zhijing Li, Yunwenyu, Tiantian Zhu, Xiao Yang and Zexuan Zhu. 2026. Enhanced Reasoning for Biomedical Document-Level Relation Extraction via a Novel Cascade Language Model Framework. 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.1517

@inproceedings{song2026enhanced,
  title = {{Enhanced Reasoning for Biomedical Document-Level Relation Extraction via a Novel Cascade Language Model Framework}},
  author = {Haohua Song, Wenhao Gu, Zhijing Li, Yunwenyu, Tiantian Zhu, Xiao Yang and Zexuan Zhu},
  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.1517},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1517}
}

References

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

  1. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, and 1 others. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774.
  2. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, and 1 others. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901.
  3. Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Baobao Chang, and 1 others. 2024. A survey on in-context learning. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 1107–1128.
  4. Abhinanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, and 1 others. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783.
  5. Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and 1 others. 2025. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. ACM Transactions on Information Systems,…
  6. Po-Ting Lai and Zhiyong Lu. 2021. Bert-gt: cross-sentence n-ary relation extraction with bert and graph transformer. Bioinformatics, 36(24):5678–5685. find
  7. Po-Ting Lai, Chih-Hsuan Wei, Ling Luo, Qingyu Chen, and Zhiyong Lu. 2023. Biorex: improving biomedical relation extraction by leveraging heterogeneous datasets. Journal of Biomedical Informatics, 146:104487. find
  8. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, and 1 others. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems, 33:9459–…
  9. 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
  10. Lishuang Li, Jing Hao, Hongbin Lu, and Jingyao Tang. 2025. Document-level biomedical relation extraction via knowledge-enhanced graph and dynamic generative adversarial networks. IEEE Transactions on Computational Biology and Bioinformatics. find
  11. Lishuang Li, Ruiyuan Lian, Hongbin Lu, and Jingyao Tang. 2022. Document-level biomedical relation extraction based on multi-dimensional fusion information and multi-granularity logical reasoning. In Proceedings of the 29th international conference on computational linguistics, pages 2098–2107. find
  12. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, and 1 others. 2024. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. find
  13. Ling Luo, Po-Ting Lai, Chih-Hsuan Wei, Cecilia N Arighi, and Zhiyong Lu. 2022. Biored: a rich biomedical relation extraction dataset. Briefings in Bioinformatics, 23(5):bbac282. find
  14. Xilai Ma, Jing Li, and Min Zhang. 2023a. Chain of thought with explicit evidence reasoning for few-shot relation extraction. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 2334–2352. find
  15. Yubo Ma, Yixin Cao, Yong Hong, and Aixin Sun. 2023b. Large language model is not a good few-shot information extractor, but a good reranker for hard samples!
  16. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 10572–10601. find
  17. OpenAI. 2025a. Introducing GPT-4.1 model family. https://openai.com/index/gpt-4-1/. Accessed: 2025-07-09. ↗
  18. OpenAI. 2025b. Introducing gpt-5. https://openai.com/index/introducing-gpt-5/. Accessed: 2025-08-07. ↗
  19. Hasin Rehana, Nur Bengisu Çam, Mert Basmaci, Jie Zheng, Christianah Jemiyo, Yongqun He, Arzucan Özgür, and Junguk Hur. 2024. Evaluating gpt and bert models for protein–protein interaction identification in biomedical text. Bioinformatics Advances, 4(1):vbae133. find
  20. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982–3992.
  21. Olaf Ronneberger, Philipp Fischer, and Thomas Brox. 2015. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention, pages 234–241. Springer.
  22. M Janina Sarol, Gibong Hong, Evan Guerra, and Halil Kilicoglu. 2024. Integrating deep learning architectures for enhanced biomedical relation extraction: a pipeline approach. Database, 2024:baae079. find
  23. Zhen Wan, Fei Cheng, Zhuoyuan Mao, Qianying Liu, Haiyue Song, Jiwei Li, and Sadao Kurohashi. 2023. Gpt-re: In-context learning for relation extraction using large language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 3534–3547.
  24. Minjia Wang, Fangzhou Liu, Xiuxing Li, Bowen Dong, Zhenyu Li, Tengyu Pan, and Jianyong Wang. 2024. Bio-rfx: refining biomedical extraction via advanced relation classification and structural constraints. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages… find
  25. Derong Xu, Wei Chen, Wenjun Peng, Chao Zhang, Tong Xu, Xiangyu Zhao, Xian Wu, Yefeng Zheng, Yang Wang, and Enhong Chen. 2024. Large language models for generative information extraction: A survey. Frontiers of Computer Science, 18(6):186357.
  26. 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 22 others. 2024. Qwen2.5 technical report. arXiv preprint arXi… find
  27. 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, pages 764–777. find
  28. Seongjun Yun, Minbyul Jeong, Raehyun Kim, Jaewoo Kang, and Hyunwoo J Kim. 2019. Graph transformer networks. Advances in neural information processing systems, 32. find
  29. 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, pages 15426–15439. find
  30. 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, pages 3999–4006. International Jo… find
  31. 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, volume 35, pages 14612–14620. find
  32. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824– 24837.
  33. Benfeng Xu, Quan Wang, Yajuan Lyu, Yong Zhu, and Zhendong Mao. 2021. Entity structure within and throughout: Modeling mention dependencies for document-level relation extraction. In Proceedings of the AAAI conference on artificial intelligence, volume 35, pages 14149–14157. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/song-2026-enhanced-reasoning-biomedical-document-level/