Published November 8, 2026 | Version 1.3.0

ODP-Reuse: a quality-gated benchmark of Ontology Design Pattern reuse

  • 1. FIZ Karlsruhe -- Leibniz Institute for Information Infrastructure
  • 2. FIZ Karlsruhe / University of Mannheim
  • 3. FIZ Karlsruhe
  • 4. FIZ Karlsruhe / Karlsruhe Institute of Technology

Description

Ontologies that verifiably reuse Ontology Design Patterns (ODPs): one unfiltered corpus of 3,705 discovered ontologies, six quality-gated views over it (679 ontologies pass the gate), an IRI-anonymised twin of the corpus, the ontology-to-pattern mapping with per-pair coverage, per-view statistics, and the complete record of how the corpus was filtered. Every ontology carries a gate verdict and, where it was excluded, the rule and the evidence for the exclusion, so any rule can be disputed and re-run rather than taken on trust. The gate removes files that are themselves copies of a pattern (which make a 100%-reuse criterion tautological), byte-identical duplicates, ontologies that contribute no entity of their own, and ontologies with no pattern overlap to score against. FILTERS.md states each rule with its threshold and counts; audit/ holds an independent re-derivation of all removals, and validate_release.py checks an unpacked copy for internal consistency. Also included: a reproducible six-stage pipeline (discover, repair, normalise, gate, audit, publish) and two worked use cases showing the corpus is consumable rather than merely descriptive -- ODP recommendation, where a recommender built on the reuse matrix improves MAP by 0.805 over what pattern documentation alone supports, and unsupervised ODP extraction, reported both against an oracle upper bound and without ever consulting the ground truth.

Notes

Version 1.3.0 adds 1,732 ontologies to the corpus from the harvest of 2026-11-08 (https://github.com/ISE-FIZKarlsruhe/ODP-Reuse/actions/runs/42466505380); 178 pass the quality gate, adding 202 ontology-pattern pairs to the views. No ontology, file or verdict of an earlier version was removed or changed. The added ontologies are listed in release-iri/additions/v1.3.0.csv.

Files

corpus-hashed.zip

Files (813.2 MB)

Name Size
md5:5309598156525db2127444daf1d1cb99
585.7 MB Preview Download
md5:b1aa994acf6eb7c145509ddc7258e7d4
221.8 MB Preview Download
md5:2c7743394e6b7881bde8aa2d1010b370
528.5 kB Download
md5:8d3a1efdbd20570c8c2c7e7e93149e8b
5.2 MB Download

Additional details