Abstract
Graph Neural Networks (GNNs) have become a powerful tool for processing graph-structured data. However, standard GNNs struggle when applied to heterophilous graphs, where connected nodes tend to have dissimilar features. In this paper, we introduce the Adaptive Neighborhood Graph Neural Network (ANGNN), a novel GNN architecture that learns to adaptively select neighbors for each node based on their feature similarity. ANGNN uses a weighted aggregation mechanism after neighborhood selection for robust representation learning on heterophilous graphs. We provide a detailed algorithm and implementation of ANGNN using PyTorch. Experimental results on several benchmark heterophilous datasets demonstrate that ANGNN significantly outperforms standard GNNs and existing state-of-the-art methods. We discuss our findings and propose future research directions.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
"Adaptive Neighborhood Graph Neural Network for Graph Learning", Technical Disclosure Commons, (February 03, 2025)
https://www.tdcommons.org/dpubs_series/7799