HybridAI-Squared at LLMs4OL 2026 Task Reuse: Divide and Conquer does not help LLMs perform Ontology Learning
Open Conference Proceedings · 02.11.2026 · 11
DOI: 10.52825/ocp.v11i.3660 ↗Kurz gefasst
Fachartikel in der Zeitschrift Open Conference Proceedings, online erschienen am 02.11.2026. Verfasst von Upal Bhattacharya und 2 weiteren Autor:innen. Die Arbeit nennt 22 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Large Language Models (LLMs) offer promising capabilities for automating ontology learning (OL) and enrichment, reducing the cost and effort associated with these tasks compared to conventional manual ontology engineering and maintenance methods. This work presents the approach and findings of the HybridAI-Squared team on the Reuse Task of the 3rd LLMs4OL Challenge at ISWC 2026. Our experimentation with direct prediction, per-relation prediction and fine-tuning showcases the effectiveness of fine-tuning which outperforms other approaches. Our results demonstrate that predicting all relation types jointly can be substantially more effective than independently predicting each relation type. Analysis indicates that per-relation prompting introduces considerable false positives, while joint prediction appears to implicitly encourage mutual exclusion among competing relation types. For per-relation approaches, precedence filtering improves performance. The improvement is derived from removing redundant relations and highlights the importance of incorporating explicit semantics when employing LLMs for OL tasks. However, less prominent and semantically complex non-taxonomic relations remain challenging for LLMs to identify. These findings demonstrate the effectiveness of fine-tuned, ontology-aware LLMs for automated OL and signify prediction of non-taxonomic relations and incorporation of external knowledge sources as promising research directions for LLM-based OL.
Zitieren
APA
Bhattacharya, U., Boer, M. D., & Sosnovsky, S. (2026). HybridAI-Squared at LLMs4OL 2026 Task Reuse: Divide and Conquer does not help LLMs perform Ontology Learning. Open Conference Proceedings, 11. https://doi.org/10.52825/ocp.v11i.3660
Vancouver
Bhattacharya U, Boer MD, Sosnovsky S. HybridAI-Squared at LLMs4OL 2026 Task Reuse: Divide and Conquer does not help LLMs perform Ontology Learning. Open Conference Proceedings. 2026;11. doi:10.52825/ocp.v11i.3660
BibTeX
@article{Bhattacharya2026,
author = {Bhattacharya, Upal and Boer, Maaike De and Sosnovsky, Sergey},
title = {HybridAI-Squared at LLMs4OL 2026 Task Reuse: Divide and Conquer does not help LLMs perform Ontology Learning},
journal = {Open Conference Proceedings},
year = {2026},
volume = {11},
doi = {10.52825/ocp.v11i.3660}
}
EUVIMED