Elamur Research⌕
☰
home / papers / 2026 / Debate to Align: Reliable Entity Alignment through Two-Stage Multi-Agent Debate

Debate to Align: Reliable Entity Alignment through Two-Stage Multi-Agent Debate

Paper · 2026 · Cunda Wang, Ziying Ma, Po Hu et al.

Summary

Entity alignment (EA) links entities across knowledge graphs (KGs) that refer to the same real-world object. LLM-based EA pipelines usually pick ambiguous entities with a heuristic similarity-gap threshold and let a single LLM choose among retrieved candidates. This leaves them limited by retrieval quality and by unchallenged one-pass reasoning. The authors propose AgentEA, which addresses both weaknesses.

First, a LLaMA-based entity encoder is fine-tuned with Direct Preference Optimization (DPO) to sharpen embeddings for candidate retrieval. Seed pairs serve as positives, and two kinds of hard negatives are used: name-similar entities and degree-aware neighbourhood entities. Second, alignment decisions pass through a two-stage multi-role debate. A cheap verification round with three agents handles uncertain entities, and only contested or low-confidence cases go on to a deeper six-agent debate.

On cross-lingual, sparse, large-scale and heterogeneous benchmarks, AgentEA outperforms KRL and LLM baselines such as ProLEA and ChatEA, often approaching saturation. This suggests that structured debate makes LLM-based alignment more reliable without depending on one strong model.

Key points

  • Training data: DPO training data come from the three cross-lingual DBP15K subsets. The ablation removing DPO drops Hits@1 on DBP15K FR-EN from 0.996 to 0.859, and on SRPRS EN-DE from 0.977 to 0.862.
  • Results on harder benchmarks: Hits@1 reaches 0.973 on ICEWS-WIKI (ProLEA: 0.964) and 0.982 on ICEWS-YAGO (ProLEA: 0.969). It also reaches 1.000 on DWY DBP-WIKI and 0.987 on DWY DBP-YAGO.
  • Debate roles: The deep stage uses attribute, alias, neighbourhood, type, attack and judge agents. Removing any single role slightly lowers performance, and replacing the debate with single-LLM reasoning hurts accuracy without proportional cost savings.
  • Robustness to settings: Performance stays stable across GPT-series and Qwen backbones of various sizes, and three debate rounds work best.
  • Limitations: The deep debate stage is computationally expensive. A correct match cannot be recovered if it is missing from the candidate set. Code is available at github.com/eryueanran/AgentEA.

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, topics and concepts

📚 Shelves 4The 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 works

Cite

Cite as Cunda Wang, Ziying Ma, Po Hu, Weihua Wang and Feilong Bao. 2026. Debate to Align: Reliable Entity Alignment through Two-Stage Multi-Agent Debate. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.297

@inproceedings{wang2026debate,
  title = {{Debate to Align: Reliable Entity Alignment through Two-Stage Multi-Agent Debate}},
  author = {Cunda Wang, Ziying Ma, Po Hu, Weihua Wang and Feilong Bao},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.297},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.297}
}

References

50 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jonas Becker. 2024. Multi-agent large language models for conversational task-solving. CoRR, abs/2410.22932. — [arXiv:2410.22932](https://arxiv.org/abs/2410.22932) find
  2. Parishad BehnamGhader, Vaibhav Adlakha, Marius Mosbach, Dzmitry Bahdanau, Nicolas Chapados, and Siva Reddy. 2024. LLM2vec: Large language models are secretly powerful text encoders. In First Conference on Language Modeling.
  3. Yixin Cao, Zhiyuan Liu, Chengjiang Li, Zhiyuan Liu, Juanzi Li, and Tat-Seng Chua. 2019. Multi-channel graph neural network for entity alignment. In Proceedings of the 57th Conference of the Association for Computational Linguistics, pages 1452–1461. find
  4. Muhao Chen, Yingtao Tian, Mohan Yang, and Carlo Zaniolo. 2017. Multilingual knowledge graph embeddings for cross-lingual knowledge alignment. In Proceedings of the International Joint Conference on Artificial Intelligence, pages 1511–1517. find
  5. Shengyuan Chen, Qinggang Zhang, Junnan Dong, Wen Hua, Qing Li, and Xiao Huang. 2024. Entity alignment with noisy annotations from large language models. In Advances in Neural Information Processing Systems. find
  6. Zerui Chen, Huiming Fan, Qianyu Wang, Tao He, Ming Liu, Heng Chang, Weijiang Yu, Ze Li, and Bing Qin. 2025. How do language models reshape entity alignment? a survey of LM-driven EA methods: Advances, benchmarks, and future. In Proceedings of the 2025 Conference on Empirical Methods in Natural Langu… find
  7. Jingwei Cheng, Chenglong Lu, Linyan Yang, Guoqing Chen, and Fu Zhang. 2025. Easyea: Large language model is all you need in entity alignment between knowledge graphs. In Findings of the Association for Computational Linguistics, pages 20981–20995. Association for Computational Linguistics. find
  8. Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, and Igor Mordatch. 2024. Improving factuality and reasoning in language models through multiagent debate. In Forty-first International Conference on Machine Learning.
  9. Zhaopeng Feng, Jiayuan Su, Jiamei Zheng, Jiahan Ren, Yan Zhang, Jian Wu, Hongwei Wang, and Zuozhu Liu. 2025. M-MAD: Multidimensional multi-agent debate for advanced machine translation evaluation. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, pages 7084–…
  10. Lingbing Guo, Zequn Sun, and Wei Hu. 2019. Learning to exploit long-term relational dependencies in knowledge graphs. In Proceedings of the 36th International Conference on Machine Learning, pages 2505–2514. find
  11. Wentao Hu, Wengyu Zhang, Yiyang Jiang, Chen Jason Zhang, Xiaoyong Wei, and Li Qing. 2025. Removal of hallucination on hallucination: Debate-augmented RAG. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, pages 15839–15853.
  12. Xuhui Jiang, Yinghan Shen, Zhichao Shi, Chengjin Xu, Wei Li, Zixuan Li, Jian Guo, Huawei Shen, and Yuanzhuo Wang. 2024a. Unlocking the power of large language models for entity alignment. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics, pages 7566–7583. Ass… find
  13. Xuhui Jiang, Yinghan Shen, Zhichao Shi, Chengjin Xu, Wei Li, Huang Zihe, Jian Guo, and Yuanzhuo Wang. 2024b. Mm-chatalign: A novel multimodal reasoning framework based on large language models for entity alignment. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 2637–… find
  14. Xuhui Jiang, Chengjin Xu, Yinghan Shen, Yuanzhuo Wang, Fenglong Su, Zhichao Shi, Fei Sun, Zixuan Li, Jian Guo, and Huawei Shen. 2024c. Toward practical entity alignment method design: Insights from new highly heterogeneous knowledge graph datasets. In Proceedings of the ACM on Web Conference 2024, p… find
  15. Dayeon Ki, Rachel Rudinger, Tianyi Zhou, and Marine Carpuat. 2025. Multiple LLM agents debate for equitable cultural alignment. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, pages 24841–24877.
  16. Guillaume Lample, Alexis Conneau, Marc’Aurelio Ranzato, Ludovic Denoyer, and Hervé Jégou. 2018. Word translation without parallel data. In 6th International Conference on Learning Representations. OpenReview.net. find
  17. Chengjiang Li, Yixin Cao, Lei Hou, Jiaxin Shi, Juanzi Li, and Tat-Seng Chua. 2019. Semi-supervised entity alignment via joint knowledge embedding model and cross-graph model. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Con… find
  18. Dongying Lin, Yinan Liu, Bin Wang, Xiaochun Yang, and 1 others. 2026. Ontotkge: Ontology-enhanced temporal knowledge graph extrapolation. arXiv preprint arXiv:2604.05468. — [arXiv:2604.05468](https://arxiv.org/abs/2604.05468) find
  19. Xiao Liu, Haoyun Hong, Xinghao Wang, Zeyi Chen, Evgeny Kharlamov, Yuxiao Dong, and Jie Tang. 2022. Selfkg: Self-supervised entity alignment in knowledge graphs. In The ACM Web Conference, pages 860–870. find
  20. Xukai Liu, Kai Zhang, Ye Liu, Enhong Chen, Zhenya Huang, Linan Yue, and Jiaxian Yan. 2023. RHGN: relation-gated heterogeneous graph network for entity alignment in knowledge graphs. In Findings of the Association for Computational Linguistics, pages 8683–8696. find
  21. Yinan Liu, Dongying Lin, Sigang Luo, Xiaochun Yang, and Bin Wang. 2026. Joint knowledge base completion and question answering by combining large language models and small language models. arXiv preprint arXiv:2604.05875.
  22. Xin Mao, Wenting Wang, Yuanbin Wu, and Man Lan. 2021. Boosting the speed of entity alignment 10 ×: Dual attention matching network with normalized hard sample mining. In The Web Conference, pages 821–832. find
  23. Xin Mao, Wenting Wang, Huimin Xu, Man Lan, and Yuanbin Wu. 2020. MRAEA: an efficient and robust entity alignment approach for cross-lingual knowledge graph. In The Thirteenth ACM International Conference on Web Search and Data Mining, pages 420–428. find
  24. Rumana Ferdous Munne, Md Mostafizur Rahman, and Yuji Matsumoto. 2025. Entity profile generation and reasoning with LLMs for entity alignment. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 20073–20086. find
  25. Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model. In Advances in Neural Information Processing Systems.
  26. Zequn Sun, Wei Hu, Qingheng Zhang, and Yuzhong Qu. 2018. Bootstrapping entity alignment with knowledge graph embedding. In Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, pages 4396–4402. find
  27. Zequn Sun, Jiacheng Huang, Wei Hu, Muhao Chen, Lingbing Guo, and Yuzhong Qu. 2019. Transedge: Translating relation-contextualized embeddings for knowledge graphs. In The International Semantic Web Conference, volume 11778, pages 612–629. find
  28. Zequn Sun, Chengming Wang, Wei Hu, Muhao Chen, Jian Dai, Wei Zhang, and Yuzhong Qu. 2020a. Knowledge graph alignment network with gated multi-hop neighborhood aggregation. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 222–229. find
  29. Zequn Sun, Qingheng Zhang, Wei Hu, Chengming Wang, Muhao Chen, Farahnaz Akrami, and Chengkai Li. 2020b. A benchmarking study of embedding-based entity alignment for knowledge graphs. Proc. VLDB Endow., 13(11):2326–2340. find
  30. Xiaobin Tang, Jing Zhang, Bo Chen, Yang Yang, Hong Chen, and Cuiping Li. 2020. BERT-INT: A bert-based interaction model for knowledge graph alignment. In Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, pages 3174–3180. find
  31. Laurens van der Maaten and Geoffrey Hinton. 2008. Visualizing data using t-sne. Journal of Machine Learning Research, 9(86):2579–2605. find
  32. Cunda Wang, Weihua Wang, Qiuyu Liang, Feilong Bao, and Guanglai Gao. 2025a. Unifying dual-space embedding for entity alignment via contrastive learning. In Proceedings of the International Conference on Computational Linguistics, pages 3110–3122. find
  33. Cunda Wang, Weihua Wang, Qiuyu Liang, and Guanglai Gao. 2025b. Local and global structure-aware contrastive framework for entity alignment. Neurocomputing, 624:129445. find
  34. Zhichun Wang, Qingsong Lv, Xiaohan Lan, and Yu Zhang. 2018. Cross-lingual knowledge graph alignment via graph convolutional networks. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 349–357.
  35. Yuting Wu, Xiao Liu, Yansong Feng, Zheng Wang, Rui Yan, and Dongyan Zhao. 2019a. Relation-aware entity alignment for heterogeneous knowledge graphs. In Proceedings of the International Joint Conference on Artificial Intelligence, pages 5278–5284. find
  36. Yuting Wu, Xiao Liu, Yansong Feng, Zheng Wang, and Dongyan Zhao. 2019b. Jointly learning entity and relation representations for entity alignment. In Proceedings of the Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Proce… find
  37. Linyan Yang, Jingwei Cheng, and Fu Zhang. 2024a. Advancing cross-lingual entity alignment with large language models: Tailored sample segmentation and zero-shot prompts. In Findings of the Association for Computational Linguistics, pages 8122–8138. find
  38. Linyao Yang, Hongyang Chen, Xiao Wang, Jing Yang, Fei-Yue Wang, and Han Liu. 2024b. Two heads are better than one: Integrating knowledge from knowledge graphs and large language models for entity alignment. CoRR, abs/2401.16960. — [arXiv:2401.16960](https://arxiv.org/abs/2401.16960) find
  39. Weixin Zeng, Xiang Zhao, Jiuyang Tang, and Xuemin Lin. 2020. Collective entity alignment via adaptive features. In 36th IEEE International Conference on Data Engineering, pages 1870–1873. find
  40. Kaiyuan Zhang, Qian Liu, Luyang Zhang, Chaoqun Zheng, Shuaimin Li, Bing Xu, Muyun Yang, Xinxiao Qiao, and Wenpeng Lu. 2025a. MADAWSD: Multi-agent debate framework for adversarial word sense disambiguation. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pag… find
  41. Taolin Zhang, Dongyang Li, Qizhou Chen, Chengyu Wang, and Xiaofeng He. 2025b. BELLE: A bi-level multi-agent reasoning framework for multi-hop question answering. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2025, Vienna, Aust…
  42. Runhao Zhao, Jiuyang Tang, Weixin Zeng, Ziyang Chen, and Xiang Zhao. 2024. Zero-shot knowledge graph question generation via multi-agent llms and small models synthesis. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, pages 3341–3351. find
  43. Runhao Zhao, Weixin Zeng, Jiuyang Tang, Yawen Li, Guanhua Ye, Junping Du, and Xiang Zhao. 2025a. Towards unsupervised entity alignment for highly heterogeneous knowledge graphs. In 2025 IEEE 41st International Conference on Data Engineering (ICDE), pages 3792–3806. IEEE. find
  44. Runhao Zhao, Weixin Zeng, Jiuyang Tang, Feng Tian, and Xiang Zhao. 2025b. Educational knowledge graph question answering: Insights from benchmarks and llm-based solutions. IEEE Transactions on Artificial Intelligence. find
  45. Runhao Zhao, Weixin Zeng, Wentao Zhang, Xiang Zhao, Jiuyang Tang, and Lei Chen. 2025c. Towards temporal knowledge graph alignment in the wild. arXiv preprint arXiv:2507.14475. — [arXiv:2507.14475](https://arxiv.org/abs/2507.14475) find
  46. Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, and Zheyan Luo. 2024. LlamaFactory: Unified efficient fine-tuning of 100+ language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations), pages 400–410.
  47. Yunfeng Zhou, Cui Zhu, Wenjun Zhu, and Hongyang Li. 2024. SCMEA: A stacked co-enhanced model for entity alignment based on multi-aspect information fusion and bidirectional contrastive learning. Neural Networks, 173:106178. find
  48. Hao Zhu, Ruobing Xie, Zhiyuan Liu, and Maosong Sun. 2017. Iterative entity alignment via joint knowledge embeddings. In Proceedings of the International Joint Conference on Artificial Intelligence, pages 4258–4264. find
  49. Qiannan Zhu, Xiaofei Zhou, Jia Wu, Jianlong Tan, and Li Guo. 2019. Neighborhood-aware attentional representation for multilingual knowledge graphs. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, pages 1943– 1949. find
  50. Xiang Zou, Xuanhong Li, Po Hu, and Ming Dong. 2025. Mrbalance: A framework for enhancing event causality identification in multi-agent debates via role assignment. Knowl. Based Syst., 330:114470. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-debate-align-reliable-entity-alignment/