Beyond Homophily in Graph Neural Networks: Current Limitations and Effective Designs

We investigate the representation power of graph neural networks in the-supervised node classification task under heterophily or low homophily,. e. , in networks where connected nodes may have different class labels and features. Many popular GNNs fail to generalize to this setting, and even outperformed by models that ignore the graph structure (e. g. , perceptrons). Motivated by this limitation, we identify a set of key -- ego- and neighbor-embedding separation, higher-order neighborhoods, combination of intermediate representations -- that boost learning from the structure under heterophily. We combine them into a graph neural network, 2GCN, which we use as the base method to empirically evaluate the of the identified designs. Going beyond the traditional with strong homophily, our empirical analysis shows that the designs increase the accuracy of GNNs by up to 40% and 27% over without them on synthetic and real networks with heterophily, , and yield competitive performance under homophily.