Reordering graphs for node embedding

Official URL: https://dx.doi.org/10.1109/SIU66497.2025.11112418

The complex connectivity patterns inherent in graphs are the biggest obstacle to applying ML algorithms, which are perfected for tabular data, to graphs. Graph embedding converts each node into a d-dimensional vector while preserving the structural graph features, thereby creating tabular structures in which each row corresponds to a node in the graph and contains d feature values. However, the embedding process is quite costly and requires accelerators such as GPUs. In this study, we propose approaches to improve the GPU-based graph embedding tools based on graph ordering. The experiments verify that reordering the graph can improve the embedding tools both in terms of speed and quality.