Justified Concept Refinement for Long-Lived Knowledge Bases: Structural Updates with Necessity, Minimality, and Identifiability Evidence

Read full text Summarise with VIRAT
Authors: Hironao Nakamura
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

When long-lived knowledge bases undergo concept refinement---splitting coarse ontological categories into finer ones in response to persistent classification failures---maintainers need formal guarantees that each change is traceable, terminating, and justified. We introduce a refinement operator for conceptual structures, grounded in the belief-change tradition (AGM) and instantiable in description logics (ALCQ), that provides such guarantees together with three modes of justification evidence: Every refinement carries a trigger_task_id yielding a lightweight why-trace, and a refinement-potential argument gives termination without depth caps. Beyond traceability and termination, we formalize three justification modes as artifact-checkable evidence objects: (i) counterfactual necessity via suppression-trial records, (ii) minimality via logged candidate sets with a declared metric and scores, and (iii) identifiability via a score-based certificate under a unique-argmin rule, with "unknown" as a first-class outcome when evidence is insufficient. An executable witness demonstrates all six properties in a fully reproducible diagnostic evaluation across 10 seeds. The result is a formal framework, with proofs of structural properties (AGM inclusion, vacuity, non-recovery; ALCQ conservative extension), that makes concept refinement in knowledge bases explicit, auditable, and justified.

Fields of Study & Subjects

Logic Reasoning and Knowledge Semantic Web and Ontologies Rough Sets and Fuzzy Logic

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
22 March 2026
Language
EN
Document type
Preprint