Bag of Tricks of Semi-Supervised Classification with Graph Neural Networks

Much of the recent progress made in node classification on graphs can be credited to the careful design on graph neural networks (GNN) and label propagation algorithms. However, in the literature, in addition to improvements to the model architecture, there are a number of improvements either briefly mentioned as implementation details or visible only in source code, and these overlooked techniques may play a pivotal role in their practical use. In this paper, we first summarize a collection of existing refinements, and then propose several novel techniques regarding these model designs and label usage. We empirically evaluate their impacts on the final model accuracy through ablation studies, and show that we are able to significantly improve various GNN models to the extent that they outweigh the gains from model architecture improvement. Notably, many of the top-ranked models on Open Graph Benchmark benefit from our techniques.