EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

VUL337-OL at LLMs4OL 2026 Tasks Flagship and Reuse: Task-Native Constraints for Ontology Construction and Extension

Chuhan Deng, Yuqi Fan, Can Zeng, Chao Zhang

Open Conference Proceedings · 02.11.2026 · 11

DOI: 10.52825/ocp.v11i.3665 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Open Conference Proceedings, online erschienen am 02.11.2026. Verfasst von Chuhan Deng und 3 weiteren Autor:innen. Die Arbeit nennt 16 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The LLMs4OL 2026 Flagship and Reuse tasks expose two complementary ontology-learning settings: constructing a primitive ontology from raw text and extending a supplied partial ontology. We study both through one design question: how should the available task structure bound a language model’s semantic decisions? Our answer is task-native constrained prediction. For Flagship, where no ontology is given, a progressive pipeline constructs and progressively narrows explicit spaces for endpoints, directed pairs, direct parents, and graph revisions while carrying evidence and stable identifiers across stages. Its quality-oriented Full configuration is complemented by a Compact revision-free operating point. For Reuse, the supplied partial ontology already localizes the problem; deterministic preprocessing derives target terms, candidate parents, and relation labels, and one structured zero-shot call selects a parent–relation attachment for every target. On the official 2,018-document Flagship test set, Full obtains 0.4285 exact edge F1, 0.4167 mean graph similarity, 0.3131 typing F1, and 0.4252 taxonomy F1; its exact edge F1 is 0.0862 higher than Compact’s. On the 1,799-record Reuse test set, the primary MiMo v2.5 Pro configuration obtains 0.9147 exact edge F1, 0.8985 semantic graph similarity, 0.9790 typing F1, and 0.9314 taxonomy F1. These results show that explicit decision spaces can support staged construction of typing and taxonomic structure from raw text and compact local extension of a supplied ontology.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Deng, C., Fan, Y., Zeng, C., & Zhang, C. (2026). VUL337-OL at LLMs4OL 2026 Tasks Flagship and Reuse: Task-Native Constraints for Ontology Construction and Extension. Open Conference Proceedings, 11. https://doi.org/10.52825/ocp.v11i.3665

Vancouver

Deng C, Fan Y, Zeng C, Zhang C. VUL337-OL at LLMs4OL 2026 Tasks Flagship and Reuse: Task-Native Constraints for Ontology Construction and Extension. Open Conference Proceedings. 2026;11. doi:10.52825/ocp.v11i.3665

BibTeX
@article{Deng2026,
  author = {Deng, Chuhan and Fan, Yuqi and Zeng, Can and Zhang, Chao},
  title = {VUL337-OL at LLMs4OL 2026 Tasks Flagship and Reuse: Task-Native Constraints for Ontology Construction and Extension},
  journal = {Open Conference Proceedings},
  year = {2026},
  volume = {11},
  doi = {10.52825/ocp.v11i.3665}
}
Kostenlos starten →