BeyondEquivalence

General description

The BeyondEquivalence track evaluates ontology matching systems on their ability to detect relations beyond equivalence. This better reflects real-world ontology alignment scenarios, where ontologies may differ in structure, granularity, and classification perspective.

Five mutually disjoint relation types are included:

Equivalence (≡) – Classes represent the exact same concept and share all instances.
Superclass_of (≥) – Every instance of the target class is also an instance of the source class (generalization).
Subclass_of (≤) – Every instance of the source class is also an instance of the target class (specialization).
Overlap (≈) – Classes share some but not all instances; their sets intersect without one being a subset of the other.
Disjoint (⊘) – Classes have no common instances; their sets are mutually exclusive.

This track addresses the need for alignment approaches capable of capturing nuanced relationships, enabling richer semantic integration for tasks such as knowledge graph merging, master data integration, and semantic search.

The results are available now.

Datasets

The benchmark dataset suite is publicly available via Zenodo:

Zenodo Record – GeneralRelation Benchmark

As part of the data is extracted from eClass, which is a non-public industrial product classification standard, usage of the eClass-derived data must comply with the terms of the eClass License.

The datasets come from three major sources: Product Classification Standards (GPC, UNSPSC, ETIM, and eClass,), and STROMA/TaSeR Test Cases [1,2]. They vary in size, structure, and semantic complexity, and all reference alignments are annotated with explicit relation types.

Data Source	Dataset	# S:Class	# T:Class	# S:Property	# T:Property	≡	≥	≤	≃
Product Classification Standards	GPC–UNSPSC	1,151	30,707	839	3	255	1,377	250	17,738
	GPC–UNSPSC⁺	5,342	30,707	839	3	381	3,306	1,293	23,600
	ETIM–eClass	2,877	5,565	10,746	8,932	2,371	380	636	291
	eClass–UNSPSC	7,349	19,600	4,009	3	329	1,533	351	55,573
	eClass–GPC	3,459	1,210	2,027	2,064	251	431	1,199	11,014
STROMA/TaSeR	g1–web	728	1,132	0	0	175	26	29	—
	g2–diseases	1,109	5,146	0	0	316	11	27	—
	g3–text	335	260	0	0	70	267	425	—
	g5–groceries	60	335	0	0	29	113	14	—
	g7–literature	41	155	0	0	12	52	18	—

Data Features

Large and heterogeneous class hierarchies with rich property structures.
Multi-source mappings combining exact matches and broader/narrower matches.
Support for n:m correspondences and partially correct alignments.
Reference alignments annotated with explicit relation types (≡, ≥, ≤, ≃, ⊘).

Contact

For inquiries or to request evaluation support, please contact the track organizers:

Xianhao Liu (xliu@stibosystems.com),

Sven Hertling (sven.hertling@uni-mannheim.de),

Michael R. Hansen (mire@dtu.dk)

Bibliography

[1] Arnold, P., Rahm, E.: Enriching ontology mappings with semantic relations. Data & Knowledge Engineering 93, 1–18 (2014)