Genuineknowledge-KEOL at LLMs4OL 2026 Tasks Flagship, Reuse, and Taxonomy: Hybrid Training-Free Pipelines for Ontology and Taxonomy Discovery
DOI:
https://doi.org/10.52825/ocp.v11i.3661Keywords:
Ontology Learning, Prompt Engineering, Large Language Models, Taxonomy DiscoveryAbstract
Ontology learning traditionally requires substantial domain expertise and curated corpora. This paper presents our systems for the three tracks of the LLMs4OL 2026 shared tasks: Task A: end-to-end ontology learning from raw text; Task B: extension of a partial ontology with new concepts and relations extracted from text; and Task C: cross-domain taxonomy learning. Our approach is entirely training-free: no model is fine-tuned, and no external knowledge bases are introduced. Rather than a single shared pipeline, the three systems follow a common design philosophy—stage-wise construction that pairs the semantic reasoning of instruction-tuned large language models, deterministic post-processing, and task-specific modules—but each task instantiates this philosophy differently. For Task A, we instantiate this framework as a domain-aware, multi-stage pipeline that retrieves in-domain demonstrations to progressively extracts and refines ontology elements. For Task B, we extend partial ontologies through domain-aware candidate verification, while using a three-model debate and ensemble mechanism to resolve conflicting concept and relation predictions. For Task C, which requires substantial domain knowledge and involves datasets of varying sizes, we employ models with stronger domain expertise together with size-adaptive chunking and rule-guided taxonomy extraction. Our work showcases how carefully orchestrated prompting, retrieval, and deterministic rules can deliver strong ontology learning results in a fully training-free setting.
Downloads
References
[1] H. Babaei Giglou, J. D'Souza, and S. Auer, "LLMs4OL: Large Language Models for Ontology Learning", in The Semantic Web – ISWC 2023, T. R. Payne, V. Presutti, G. Qi et al., Eds., Cham: Springer Nature Switzerland, 2023, pp. 408–427. ISBN: 978-3-031-47240-4.
[2] H. B. Giglou, J. D'Souza, and S. Auer, "LLMs4OL 2024 Overview: The 1st Large Language Models for Ontology Learning Challenge", Open Conf. Proc., vol. 4, pp. 3–16, Oct. 2024. DOI: 10.52825/ocp.v4i.2473. [Online]. Available: https://www.tib-op.org/ojs/index.php/ocp/article/view/2473.
[3] H. Babaei Giglou, J. D'Souza, S. Sadruddin, and S. Auer, "LLMs4OL 2024 Datasets: Toward Ontology Learning with Large Language Models", Open Conf. Proc., vol. 4, pp. 17–30, 2024.
[4] H. B. Giglou, J. D'Souza, N. Mihindukulasooriya, and S. Auer, "LLMs4OL 2025 Overview: The 2nd Large Language Models for Ontology Learning Challenge", Open Conf. Proc., vol. 6, Oct. 2025. DOI: 10.52825/ocp.v6i.2913. [Online]. Available: https://www.tib-op.org/ojs/index.php/ocp/article/view/2913.
[5] H. Babaei Giglou, J. D'Souza, A. Aioanei, N. Mihindukulasooriya, and S. Auer, OntoLearner: A Modular Python Library for Ontology Learning with Large Language Models, 2026. [Online]. Available: https://arxiv.org/abs/2607.01977, arXiv: 2607.01977 [cs.AI].
[6] LLMs4OL 2026 organisers, "The 3rd LLMs4OL Challenge @ ISWC 2026 — Challenge Overview", 2026. Accessed: Jul. 19, 2026. [Online]. Available: https://sites.google.com/view/llms4ol2026/challenge-overview.
[7] R. Rahnamoun and M. Shamsfard, "SBU-NLP at LLMs4OL 2025 Tasks A, B, and C: Stage-wise Ontology Construction through LLMs without Any Training Procedure", Open Conf. Proc., vol. 6, 2025.
[8] Anthropic, "Claude Opus 4.6", 2026. Accessed: Jul. 16, 2026. [Online]. Available: https://docs.anthropic.com/en/docs/about-claude/models/overview.
[9] Anthropic, "Claude Opus 4.7", 2026. Accessed: Jul. 16, 2026. [Online]. Available: https://docs.anthropic.com/en/docs/about-claude/models/overview.
[10] DeepSeek-AI, DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence, 2026. [Online]. Available: https://arxiv.org/abs/2606.19348, arXiv: 2606.19348.
[11] Y. Zhang, M. Li, D. Long et al., Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models, 2025. [Online]. Available: https://arxiv.org/abs/2506.05176, arXiv: 2506.05176 [cs.CL].
[12] N. Muennighoff, N. Tazi, L. Magne, and N. Reimers, "MTEB: Massive Text Embedding Benchmark", in Proc. of the 17th Conf. of the European Chapter of the Assoc. for Computational Linguistics (EACL), 2023, pp. 2014–2037. [Online]. Available: https://arxiv.org/abs/2210.07316.
[13] T. B. Brown, B. Mann, N. Ryder et al., "Language Models are Few-Shot Learners", in Advances in Neural Information Processing Systems, vol. 33, 2020, pp. 1877–1901.
[14] Y. Du, S. Li, A. Torralba, J. B. Tenenbaum, and I. Mordatch, "Improving Factuality and Reasoning in Language Models through Multiagent Debate", in Proc. of the 41st Int. Conf. on Machine Learning (ICML), 2024. [Online]. Available: https://arxiv.org/abs/2305.14325.
[15] T. Liang, Z. He, W. Jiao et al., "Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate", in Proc. of the 2024 Conf. on Empirical Methods in Natural Language Processing (EMNLP), 2024. [Online]. Available: https://arxiv.org/abs/2305.19118.
[16] Anthropic, "Prompt Caching", 2026. Accessed: Jul. 16, 2026. [Online]. Available: https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching.
[17] Sentence-Transformers, "all-MiniLM-L6-v2: A Distilled Sentence-Embedding Model", 2021. Accessed: Jul. 16, 2026. [Online]. Available: https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2.
[18] LLMs4OL 2026 organisers, "LLMs4OL 2026 — Flagship Task", 2026. Accessed: Jul. 27, 2026. [Online]. Available: https://sites.google.com/view/llms4ol2026/flagship-task.
[19] LLMs4OL Challenge, "LLMs4OL Challenge 2026: Evaluation Metrics", 2026. Accessed: Jul. 27, 2026. [Online]. Available: https://github.com/sciknoworg/LLMs4OL-Challenge/tree/main/2026/metrics.
Downloads
Published
How to Cite
Conference Proceedings Volume
Section
License
Copyright (c) 2026 Lvting Liu, Yubin Luo, Ji Qi, Chun Zheng, Chengli Zhong, Yi Zhou

This work is licensed under a Creative Commons Attribution 4.0 International License.