Skip to content
Conference Open access

Volt-IQ at LLMs4OL 2026 Tasks Flagship, Reuse, and Taxonomy: Constrained Graph Synthesis and Domain-Calibrated Ontology Learning

Nov 2026 · Open Conference Proceedings · 0 citations · 25 references

Abstract

Large language models can generate ontology elements, but the evidence and output constraints differ across document extraction, graph extension, and taxonomy induction. This paper presents the Volt-IQ systems for LLMs4OL 2026 Tasks Flagship, Reuse, and Taxonomy. A shared processing contract routes each task to a different candidate source. Task A uses a merged LLM graph-extraction artifact; Task B combines normalized training memory with record-local direct-parent recovery from public ontologies; and Task C selects lexical or ontology-backed taxonomy induction by domain. In organizer intermediate evaluation, the best retained Task A run obtained exact, fuzzy, and semantic graph similarity of 0.5484, 0.6937, and 0.7715. The corresponding Task B values were 0.2253, 0.3659, and 0.4811. Best exact F1 for the four Task C domains ranged from 0.1138 to 0.1914. All six final JSON artifacts passed repository schema validation, but hidden-final scores were unavailable. The evidence therefore supports a qualified engineering conclusion: task-specific constraints provide a practical way to control multi-task ontology construction, while public-ontology overlap, target-domain calibration, incomplete ontology versioning, and gaps in final-artifact provenance limit generalization and reproducibility claims.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.