Abstract
Graph Neural Networks (GNNs) have shown significant success in learning from graph-structured data. However, their performance often degrades on heterophilous graphs, where connected nodes tend to have dissimilar features. This paper introduces the Disentangled Graph Learning Network (DGLN), a novel GNN model designed specifically to learn on heterophilous networks by disentangling node features into structural and content components. DGLN uses an adaptive aggregation module and a graph reconstruction module to learn node representations and model network structure. We provide a detailed description of DGLN, including its algorithm, and implementation with PyTorch. Experimental results on several benchmark heterophilous datasets show that DGLN significantly outperforms standard GNNs and state-of-the-art methods. We discuss the implications of our results and outline future research directions.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
"Disentangled Graph Learning", Technical Disclosure Commons, (March 10, 2025)
https://www.tdcommons.org/dpubs_series/7886