KG-SaF-Data: Complete KGR Datasets

KG-SaF-Data provides complete and curated knowledge graph datasets designed for machine learning and reasoning tasks. Each dataset comes with both a schema (TBox) and instance data (ABox), along with role definitions (RBox), making them ready for Knowledge Graph Refinement (KGR) research and embedding-based pipelines.

These datasets are compatible with Python, PyTorch, PyKEEN, and ontology editors like Protege.

Dataset Structure

KG-SaF datasets follow a Description Logic (DL) formalization, organizing the knowledge graph into three main components:

• ABox (Assertional Box): Instance-level data (entities, class assertions, object property assertions)

• TBox (Terminological Box): Schema-level knowledge, including class hierarchies and axioms

• RBox (Role Box): Relationships and properties, including domain, range, and hierarchy

Note

📄 Files marked with this icon are new serializations or variations of the same data already available in OWL format (e.g., TSV or JSON representations), intended for easier use in ML pipelines.

Each dataset folder contains the following structure:

📁 abox ......................................... # Assertional Box (instance-level data)
│ ├── 📁 splits ................................. # Train/test/validation splits
│ │ ├── 🦉 train.nt ............................. # Training triples (N-Triples)
│ │ ├── 🦉 valid.nt ............................. # Validation triples (N-Triples)
│ │ ├── 🦉 test.nt .............................. # Test triples (N-Triples)
│ │ ├── 📄 train.tsv ............................ # Training triples (TSV)
│ │ ├── 📄 valid.tsv ............................ # Validation triples (TSV)
│ │ └── 📄 test.tsv ............................. # Test triples (TSV)
│ │
│ ├── 🦉 individuals.owl ........................ # Individuals definitions
│ ├── 🦉 class_assertions.owl ................... # Individuals class assertions (OWL)
│ ├── 📄 class_assertions.json .................. # Individuals class assertions (JSON)
│ │
│ ├── 🦉 obj_prop_assertions.nt ................. # Combined triples (N-Triples)
│ └── 📄 obj_prop_assertions.tsv ................ # Combined triples (TSV)

📁 rbox ......................................... # Role Box (relations and properties)
│ ├── 🦉 roles.owl .............................. # Role definitions
│ ├── 📄 roles_domain_range.json ................ # Domain and range of roles (JSON)
│ └── 📄 roles_hierarchy.json ................... # Role hierarchy (JSON)

📁 tbox ......................................... # Terminological Box (schema-level info)
│ ├── 🦉 classes.owl ............................ # Class non-taxonomical axioms
│ ├── 🦉 taxonomy.owl ........................... # Hierarchical taxonomy
│ └── 📄 taxonomy.json .......................... # Hierarchical taxonomy (JSON)

🦉 knowledge_graph.owl .......................... # Full merged TBox + RBox + ABox
🦉 ontology.owl ................................. # Core modularized schema

📁 mappings ..................................... # Mappings to IDs
│ ├── 🧾 class_to_id.json ....................... # Map ontology classes to IDs
│ ├── 🧾 individual_to_id.json .................. # Map entities/instances to IDs
│ └── 🧾 object_property_to_id.json ............. # Map object properties to IDs

Available Ontologies

KG-SaF includes datasets derived from well-known knowledge graphs and domain-specific ontologies. Each ontology is provided as part of the dataset with modularized TBox, RBox, and ABox components.

• DBpedia – Link
  Large-scale multilingual knowledge graph extracted from Wikipedia. Contains general-purpose concepts like Person, Place, Organisation, and relationships between them.

• YAGO3 – Link
  Knowledge graph integrating Wikipedia, WordNet, and GeoNames. Known for high coverage and rich semantic types (ALHIF+ DL fragment).

• YAGO4 – Link
  Updated version of YAGO with enhanced temporal and factual coverage, suitable for reasoning over large-scale datasets.

• ArCo – Link
  Cultural heritage ontology and KG covering Italian assets. Supports advanced reasoning with SROIQ DL fragment.

• WHOW – Link
  Ontology for the World Heritage and cultural objects, designed for semantic reasoning and ML experiments.

• ApuliaTravel – Link
  Domain-specific KG for tourism in Apulia, Italy. Models attractions, accommodations, and travel routes.

Available Datasets

The table below lists the currently available ontologies and their corresponding datasets included in this resource.

Note

This table will be updated as new datasets and ontologies become available.

Ontology	Dataset	DL Fragment
DBpedia	DBPEDIA25-50K-C	ALCHF
DBpedia	DBPEDIA25-100K-C	ALCHF
YAGO3	YAGO3-39K-C	ALHIF+
YAGO3	YAGO3-10-C	ALHIF+
YAGO4	YAGO4-20-C	ALCHIF
ArCo	ARCO25-20	SROIQ
ArCo	ARCO25-10	SROIQ
ArCo	ARCO25-5	SROIQ
WHOW	WHOW25-5	SROIQ
ApuliaTravel	ATRAVEL	SRIQ

Dataset Statistics

Table: Available datasets schema axiom coverage comparison

Legend:

• ✅ = axiom available in the dataset

• ❌ = axiom not available

• ⚠️ = axiom available in the KG but not in the dataset

Axiom Type	DB100K-C	YAGO3-10-C	YAGO4-20-C	APULIA	WHOW-5	ARCO-5
ClassAssertion	✅	✅	✅	✅	✅	✅
SubClassOf	✅	✅	✅	✅	✅	✅
EquivalentClasses	❌	❌	❌	✅	❌	✅
DisjointClasses	✅	❌	✅	✅	✅	✅
UnionOf	❌	❌	✅	✅	✅	✅
IntersectionOf	❌	❌	❌	❌	❌	❌
ComplementOf	❌	❌	❌	✅	❌	✅
Existential Restrictions	❌	❌	❌	✅	✅	✅
Universal Restrictions	❌	❌	❌	✅	✅	✅
Cardinality Restrictions	❌	❌	❌	✅	✅	✅
ObjPropDomain	✅	✅	✅	✅	✅	✅
ObjPropRange	✅	✅	✅	✅	✅	✅
SubObjProp	✅	✅	✅	✅	✅	✅
InverseObjProp	❌	❌	✅	✅	✅	✅
EquivalentObjProp	⚠️	❌	❌	✅	✅	✅
ObjPropCharacteristic	⚠️	✅	✅	✅	✅	✅
ObjPropChain	❌	❌	✅	✅	✅	✅

Table: Available datasets statistics

Dataset statistics including ABox (instance-level) and Schema/TBox (class & property-level) information. For object property structural columns, the highest values in each column are highlighted in bold (manually indicated with …).

Dataset	Triples	Inds	Props	Classes	1to1	1toN	Nto1	NtoN	Avg Triples	Class Assert.	TBox Classes	Disjoints	Subclass	∃R.C	∀R.C	Props	Domain	Range	Both	Functional
DB-50K-C	28,525	22,268	275	169	0.21	0.08	0.33	0.38	103.73	12,419	232	14	226	-	-	280	194	217	152	-
DB-100K-C	577,249	96,375	406	229	0.09	0.06	0.15	0.71	1,421.80	82,750	297	14	290	-	-	411	295	317	233	-
Y3-10-C	1,080,398	123,038	34	92,539	0.00	0.00	0.12	0.88	31,776.41	1,309,964	94,726	-	94,809	-	-	35	34	29	28	12
Y3-39K-C	370,169	37,711	34	45,456	0.03	0.03	0.15	0.79	10,887.32	-	46,892	-	46,960	-	-	35	34	29	28	10
Y4-20-C	653,988	91,904	68	1,433	0.07	0.07	0.13	0.72	9,617.47	116,141	1,462	9	1,665	-	-	69	68	56	56	6
ATRAVEL	76,943	29,767	25	53	0.36	0.16	0.36	0.12	3,077.72	35,910	100	12	184	17	52	71	57	61	47	16
WHOW-5	584,791	137,740	25	31	0.00	0.00	0.76	0.24	23,391.64	144,892	91	16	202	43	34	102	99	99	99	-
ARCO-20	95,840	15,690	53	78	0.04	0.09	0.40	0.47	1,808.30	41,626	315	14	873	181	366	454	348	352	271	62
ARCO-10	202,492	45,400	111	117	0.06	0.06	0.44	0.44	1,824.25	119,304	378	24	1,033	211	431	593	462	465	369	71
ARCO-5	655,089	198,674	196	192	0.10	0.10	0.40	0.40	3,342.29	471,031	438	26	1,153	228	474	683	546	551	450	84