[Paper Review] pyRDF2Vec: A Python Implementation and Extension of RDF2Vec

Gilles Vandewiele, Bram Steenwinckel+2 more|arXiv (Cornell University)|May 4, 2022

Computational Physics and Python Applications4 citations

TL;DR

This paper introduces pyRDF2Vec, a Python-ported and extended implementation of the RDF2Vec algorithm for learning knowledge graph embeddings. It enables data scientists to easily integrate RDF2Vec into ML pipelines by providing optimized walk extraction, bundled extensions, and production-grade tooling, significantly improving usability and extensibility over the original Java-based version.

ABSTRACT

This paper introduces pyRDF2Vec, a Python software package that reimplements the well-known RDF2Vec algorithm along with several of its extensions. By making the algorithm available in the most popular data science language, and by bundling all extensions into a single place, the use of RDF2Vec is simplified for data scientists. The package is released under a MIT license and structured in such a way to foster further research into sampling, walking, and embedding strategies, which are vital components of the RDF2Vec algorithm. Several optimisations have been implemented in exttt{pyRDF2Vec} that allow for more efficient walk extraction than the original algorithm. Furthermore, best practices in terms of code styling, testing, and documentation were applied such that the package is future-proof as well as to facilitate external contributions.

Motivation & Objective

• Address the limited adoption of RDF2Vec in data science workflows due to its original Java implementation, which is less popular than Python.
• Simplify the integration of RDF2Vec into data science pipelines by providing a Python-native, well-documented, and extensible software package.
• Unify multiple existing RDF2Vec extensions into a single, cohesive implementation to facilitate experimentation and future research.
• Improve performance and scalability for large knowledge graphs through optimized walk extraction mechanisms.
• Enable long-term maintainability and community contributions via automated testing, continuous integration, and comprehensive documentation.

Proposed method

• Reimplement the original RDF2Vec algorithm in Python to align with the dominant data science ecosystem.
• Integrate multiple published extensions of RDF2Vec—such as context-aware walk sampling and entity type-aware walks—into a single, modular codebase.
• Optimize walk extraction using efficient graph traversal techniques and data structures to improve performance over the original Java implementation.
• Employ modern software engineering practices: continuous integration (CI), continuous delivery (CD), continuous testing (CT), and static type checking via mypy.
• Use Poetry for dependency management and Sphinx for automated, versioned documentation hosted on Read the Docs.
• Apply standardized code styling and comprehensive unit testing with pytest to ensure code quality and facilitate external contributions.

Experimental results

Research questions

• RQ1 How can the RDF2Vec algorithm be effectively reimplemented in Python to improve accessibility for data scientists?
• RQ2 To what extent can performance and scalability be improved in walk extraction through algorithmic and architectural optimizations?
• RQ3 Can bundling multiple RDF2Vec extensions into a single package enhance research reproducibility and foster innovation in embedding strategies?
• RQ4 How effective is the adoption of modern software engineering practices in ensuring long-term maintainability and community contribution?
• RQ5 What is the real-world impact and usage of pyRDF2Vec in downstream research and applications?

Key findings

• pyRDF2Vec has achieved 17,500 downloads and 146 GitHub stars, indicating strong adoption in the research community.
• The package has been used in 31 published studies on Google Scholar, demonstrating its integration into diverse research domains such as link prediction, semantic similarity, and explainable AI.
• Several studies have used pyRDF2Vec as a baseline or comparison method, including in tasks like news tag recommendation, protein similarity modeling, and target-disease link prediction.
• The use of continuous integration, testing, and documentation practices has led to increased stability, reliability, and contributor accessibility.
• The adoption of Poetry and Sphinx has streamlined dependency management and automated documentation updates, enhancing long-term maintainability.
• The performance improvements in walk extraction make pyRDF2Vec suitable for large-scale knowledge graphs, enabling efficient embedding learning in real-world data science pipelines.