Summary
Graph neural networks (GNNs) are a type of AI that work well when connected things have similar labels but struggle when connected things are different, a situation called heterophily. The authors studied how changing the way nodes are described (their features) affects the performance of different GNN designs. They found that some network designs gain much more from better node descriptions than others, showing that the choice of network and node features interact closely. However, the overall ranking of which networks work better stays mostly the same, suggesting the main difference is in how sensitive each design is to changes in node details.
What this means in practice
- •For machine learning engineers: Select graph neural network architectures based on their sensitivity to improved node feature representations to optimize heterophilic graph performance.
- •For natural language processing developers: Design node feature embeddings using contextual Transformers to enhance graph learning tasks involving text in product or historical data graphs.
Abstract
Graph Neural Networks (GNNs) perform well on homophilic graphs but struggle in heterophilic settings, where connected nodes often carry dissimilar labels. Existing evaluations typically compare architectures under a fixed node-feature representation, leaving unclear whether conclusions about heterophily robustness remain stable as the input representation changes. We address this question by constructing parallel feature variants of two large-scale heterophilic benchmarks, Roman-Empire and Amazon-Ratings, pairing each graph with representations ranging from static fastText vectors to contextual Transformer embeddings and evaluating seven GNN architectures across these representations. We find that the effect of representation varies across architectures: on Roman-Empire, the contextual gain ranges from 2.38 percentage points for GCN-sep to 13.67 points for GAT, with H2GCN gaining 8.77 points. On Amazon-Ratings, where node text is limited to short product titles, GAT improves by 6.78 points from fastText to MPNet, while GCN-sep changes by only 0.20 points. These results show that architectural performance is conditional on node representation: the same representation change can produce different magnitudes of performance gain across architectures, so architecture and representation cannot be treated as independent evaluation factors. A rank-correlation analysis on these two benchmarks further shows that the relative ordering of architectures remains highly stable across representations, isolating differential sensitivity, rather than ranking instability, as the primary effect.