Edge Entropy as an Indicator of the Effectiveness of GNNs over CNNs for Node Classification
This work provides a metric for researchers to predict the effectiveness of GNNs over CNNs for node classification, helping to guide model selection.
The authors introduce edge entropy to quantify the performance improvement of Graph Neural Networks (GNNs) over Convolutional Neural Networks (CNNs) for node classification. They found that lower edge entropy values predict larger performance gains for GNNs, while higher edge entropy leads to smaller improvements.
Graph neural networks (GNNs) extend convolutional neural networks (CNNs) to graph-based data. A question that arises is how much performance improvement does the underlying graph structure in the GNN provide over the CNN (that ignores this graph structure). To address this question, we introduce edge entropy and evaluate how good an indicator it is for possible performance improvement of GNNs over CNNs. Our results on node classification with synthetic and real datasets show that lower values of edge entropy predict larger expected performance gains of GNNs over CNNs, and, conversely, higher edge entropy leads to expected smaller improvement gains.