Toolboxes & Code Repositories

RR-GCN

Library for creating relational graph convolutional Network embeddings 

software
Description:

RR-GCN is a simplified graph representation method that leverages the "message passing" paradigm to construct node embeddings through random transformations, proving that structural aggregation often outweighs learned weights in Knowledge Graphs. RR-GCN is built to be efficient; it supports node classification tasks by aggregating randomly transformed representations from neighbors, allowing it to compete with fully trained R-GCNs while significantly reducing training overhead.

More Information:

  • Originates from the AI-Driven Data Science research challenge (FAIR 2019–2023).

  • Responsible Research Lead: Femke Ongenae

  • Authors: Sandeep Ramachandra, Vic Degraeve, Gilles Vandewiele, Bram Steenwinckel, Sofie Van Hoecke, Femke Ongenae

  • Programming Language: Python

  • Link to the repository

  • Categories: Networks and graphs

Publications:

  • Ramachandra, S., Degraeve, V., Vandewiele, G., Steenwinckel, B., Van Hoecke, S., & Ongenae, F. (2025). RR-GCN : exploring untrained random embeddings for relational graphs. INTERNATIONAL JOURNAL OF SOFTWARE ENGINEERING AND KNOWLEDGE ENGINEERING, 35(6), 809–834. https://doi.org/10.1142/S0218194025500184