Metraipolis › archive › dc_4de515f05e47
SYNTHOLOGY: Ontology-Based Data Generation for Neurosymbolic Reasoning (2026) · DOI 10.3233/ssw260018
- Source
- Metraipolis archive
- Original
- https://doi.org/10.3233/ssw260018
- License (as archived)
- CC-BY-4.0
- Archived
- 2026-10-08T03:45:44.291Z
- SHA-256 of the archived text
456d2f78289b6b22035ba97f5142b145cdda2733e890e5ce5d3887b36deadf7b- Stable URL of this page
https://metraipolis.com/documento/dc_4de515f05e47
How to cite
SYNTHOLOGY: Ontology-Based Data Generation for Neurosymbolic Reasoning (2026) · DOI 10.3233/ssw260018 — Metraipolis archive, https://doi.org/10.3233/ssw260018 (CC-BY-4.0); archived copy: https://metraipolis.com/documento/dc_4de515f05e47 (Metraipolis archive dc_4de515f05e47, sha256 456d2f78289b6b22035ba97f5142b145cdda2733e890e5ce5d3887b36deadf7b)
BibTeX
@misc{dc4de515f05e47,
title = {SYNTHOLOGY: Ontology-Based Data Generation for Neurosymbolic Reasoning (2026) · DOI 10.3233/ssw260018},
howpublished = {Metraipolis archive},
url = {https://doi.org/10.3233/ssw260018},
note = {License CC-BY-4.0; archived copy https://metraipolis.com/documento/dc_4de515f05e47; sha256 456d2f78289b6b22035ba97f5142b145cdda2733e890e5ce5d3887b36deadf7b},
year = {2026}
}Archived text
Reproduced under CC-BY-4.0. Attribution: Metraipolis archive, https://doi.org/10.3233/ssw260018. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Articolo recente con licenza aperta: scheda dai metadati Crossref (book-chapter). Titolo: SYNTHOLOGY: Ontology-Based Data Generation for Neurosymbolic Reasoning Autori: Vincent Van Schependom, Cas Proost, Pieter Bonte Pubblicato: 2026-09-02 · in: Studies on the Semantic Web · editore: IOS Press DOI: https://doi.org/10.3233/ssw260018 Licenza dell'articolo: CC-BY-4.0 Abstract (testo degli autori, CC-BY-4.0): Symbolic reasoners offer a simple contract: provide an ontology, receive entailments. Neurosymbolic reasoners promise the same task with added scalability and tolerance to noise, but break this contract by requiring an ontology-specific training dataset, which in most cases does not exist and must be gathered from scratch. This bottleneck blocks practitioners from applying neurosymbolic reasoners to new ontologies. We present SYNTHOLOGY, an ontology-agnostic synthetic data generator that closes this gap by producing the required training data from any OWL 2 RL ontology, with no need for manual data gathering. SYNTHOLOGY engineers data through ontology-guided backward-chaining proof search, which guarantees deep multi-hop derivations by construction, and introduces proof-based negative sampling. Together, these properties force models to traverse the full reasoning chain to correctly classify triples. Our evaluation shows that a model trained on SYNTHOLOGY data outperforms models trained on baseline data; part of this advantage reflects test-set familiarity, though the gain in rejecting near-miss negatives persists under an independently generated test set (Section 6.2). Proof-based negative sampling yields consistent gains over random and constrained sampling on all learning-quality metrics, and SYNTHOLOGY scales to ontologies on which traditional (ontology specific) generators run out of memory. To our knowledge, SYNTHOLOGY is the first ontology-agnostic synthetic data generator that provides multi-hop coverage, near-miss negatives, and derivation metadata by construction, a step toward making neurosymbolic reasoners as readily deployable on a new ontology as their symbolic counterparts. Fonte: https://api.crossref.org/works/10.3233%2Fssw260018 — metadati Crossref, consultati il 2026-10-07. Interrogazione: «linked data semantic web».