AIDOC-AP - AI Documentation Application Profile

# AIDOC-AP: Application Profile for Technical Documentation of AI Systems **Version:** 1.2 (July 2026) **License:** CC BY 4.0 **Project:** CERTAIN (Horizon Europe Grant Agreement No. 101189650) ## Authors - Sebastian Neumaier (University of Applied Sciences St. Pölten & Vienna University of Economics and Business, Austria)- Tobias Dam (University of Applied Sciences St. Pölten, Austria)- Fabian Kovac (University of Applied Sciences St. Pölten, Austria) ## Description AIDOC-AP is an ontology-based application profile for representing the technical documentation of AI systems required under Article 11 and Annex IV of the EU AI Act in a structured, machine-readable form. It reuses established Semantic Web vocabularies (PROV-O, DCAT, MLSchema, AIRO, VAIR, DPV) and adds Annex IV-specific terms, so that provider documentation can be captured as a knowledge graph and checked for completeness against the regulation. Version 1.2 models categorical concepts as classes instead of string-valued properties (IntendedPurpose, Modality), generalises VisualDocumentation to TechnicalDocumentation, grounds AIDOC-AP terms in the Data Privacy Vocabulary (DPV v2.3, incl. its AI, TECH and AI Act extensions) via rdfs:subClassOf and SKOS mapping relations, and publishes expert-curated alignments to reference ontologies together with the full evaluation data (coverage experiments over four open-weight LLMs, SPARQL-based competency-question validation, curation and expert-agreement records). ## Core Ontology Files 1. **aidoc-ap.ttl** — Main ontology (v1.2) defining classes for AI systems, models, datasets, and documentation artifacts; reuses PROV-O, DCAT, MLSchema, AIRO, VAIR, and DQV vocabularies and is grounded in DPV 2. **annex_4.ttl** — Formal representation of the 22 Annex IV requirements using SKOS concept schemes, with lifecycle annotations and 50 competency questions 3. **aidoc-ap-v1.2.zip** — Full repository snapshot at release v1.2: five pilot knowledge graphs (energy, biometrics, finance, civic participation, HR), the 50 competency questions as executable SPARQL queries, the expert-curated alignments, the published experiment data, and the complete analysis toolchain ## Web Access - **Persistent URI:** https://w3id.org/aidoc-ap- **GitHub Repository:** https://github.com/CERTAIN-Project/aidoc-ap- **Documentation:** https://certain-project.github.io/aidoc-ap/