KG-SaF: Data and Workflow Documentation

KG-SaF-Data and KG-SaF-JDeX from “Diliso, I., Barile, R., d’Amato, C., & Fanizzi, N. (2025). KG-SaF: Building Complete and Curated Datasets for Machine Learning and Reasoning on Knowledge Graphs (Version 0.0.0.1) [Computer software]. https://doi.org/10.5281/zenodo.17817931”

KG-SaF provides a workflow (KG-SaF-JDeX) and curated datasets (KG-SaF-Data) for knowledge graph refinement (KGR) research. The resource includes datasets with both schema (ontologies) and ground facts, making it ready for machine learning and reasoning services.

Key Features

• 🗂️ Extracts datasets from RDF-based KGs with expressive schemas (RDFS/OWL2)

• 📦 Provides datasets in OWL and TSV formats, easily loadable in both PyTorch and Protege

• ⚡ Handles inconsistencies and leverages reasoning to infer implicit knowledge

• 🤖 Provides ML-ready tensor representations compatible with PyTorch and PyKEEN

• 🧩 Offers schema decomposition into themed partitions (modularization of ontology components)

Contents:

• Quick Start Guide
  • Software Requirements
  • Installation Instructions
  • System Requirements
  • Unpack Released Datasets
• Tutorials and Guides
  • Loading and Analyzing a KG-SaF Datasets in PyTorch
  • Running TransE on KG-SaF Datasets
  • Using KG-SaF-JDeX on Custom Knowledge Graphs
• KG-SaF-Data: Complete KGR Datasets
  • Dataset Structure
  • Available Ontologies
  • Available Datasets
  • Dataset Statistics
• KG-SaF-JDeX: Modular Dataset Generation Workflow
  • Machine Learning Loaders
  • Reasoning Utilities
  • Modularization and Decomposition
  • Conversion Utilities

Appendix:

• References