H₂KG Ontology

H₂KG-CHAMEO Characterization Benchmark

Domain-specific ontology for hydrogen electrochemical systems, FAIR workflow representation, and external semantic alignment

H₂KG Ontology logoOntology release

Developed through DECODE, the Helmholtz Metadata Collaboration (HMC), and AIMWORKS to support FAIR ontology publication and reusable metadata infrastructure for hydrogen electrochemical systems.

Controlled ontology application benchmark

One source corpus. Two native semantic models.

H₂KG and CHAMEO were mapped independently from the same sanitized characterization manifest. It contains 8 source records across 7 method families and 7 unique profiles. The frozen requirements, expected rows, and conclusion rules were fixed before query retrieval or native-pattern assessment. Neither ontology was modified, and no local semantic bridge was admitted to either scored graph.

H₂KG exact12/1392.3%
CHAMEO exact10/1376.9%
Expected rows501across eight source records
Import closure25CHAMEO/EMMO modules

Fairness controls

How the comparison was kept symmetric

01

Common source

Both ABoxes begin with the same 1,635 eligible fields from eight records representing seven unique profiles.

02

Independent mapping

Each graph uses its own ontology and declared imports; neither graph is translated from the other.

03

Mixed assessment

Thirteen lifecycle requirements use query retrieval, native-pattern assessment, or native vocabulary coverage.

04

Auditable decision

Five hashed rules must all pass before the complementarity conclusion is emitted.

Computed outcome

For the selected characterization profiles, the benchmark supports a scoped complementarity conclusion: CHAMEO provides explicit characterization-methodology structures, while H₂KG provides greater native class coverage at the predefined profile granularity together with direct valued-parameter and individually provenance-linked result patterns.

The 12/13 and 10/13 values are requirement-satisfaction rates under this frozen protocol, not general ontology-accuracy or quality scores. The benchmark covers eight source records representing seven unique profiles; governance, maintainability, reasoning performance, usability, and scalability were not measured.

Quantity interpretation

Parsing is separate from scientific compatibility

The result-value requirement contains 13 records with a parseable number and recognized unit. A separate rule-based screen found 11 records without a detected property/unit conflict and 2 conflicts. The two conflicts are the duplicated source entry electrolyte saturation = 47 mol/L: mol/L denotes concentration, whereas saturation is ordinarily dimensionless. The raw entries remain queryable but are not treated as scientifically compatible or independently verified saturation quantities.

Query and structural evidence

Trace every benchmark outcome

Select a plain-language requirement and ontology. Query-based requirements show frozen RDFLib output; native-pattern and vocabulary assessments show their recorded evidence and rationale.

Loading benchmark package...

Scientific question

Observed

SPARQL result rows

Expected

Manifest-derived ground truth

Query evidence subgraph

Why this row is returned

Choose a row to see its compact RDF support. Solid arrows are asserted predicates; absent required paths are reported as gaps rather than invented.

WorkflowProcessMeasurementMatterParameterData/resultPropertyInstrument/softwareValue/unit

Normalized, deduplicated set comparison

Missing and additional answers