Scalable Graph Neural Networks for Heterogeneous Graphs

TL;DR

This paper introduces NARS, a scalable neighbor averaging method for heterogeneous graphs that achieves state-of-the-art accuracy by efficiently utilizing relation subgraphs without complex end-to-end training.

Contribution

The paper presents NARS, a novel scalable approach for heterogeneous graphs that outperforms existing GNN methods by using neighbor averaging over relation subgraphs.

Findings

NARS achieves state-of-the-art accuracy on benchmark datasets.

NARS is more memory-efficient during training and inference.

NARS outperforms more expensive GNN-based methods.

Abstract

Graph neural networks (GNNs) are a popular class of parametric model for learning over graph-structured data. Recent work has argued that GNNs primarily use the graph for feature smoothing, and have shown competitive results on benchmark tasks by simply operating on graph-smoothed node features, rather than using end-to-end learned feature hierarchies that are challenging to scale to large graphs. In this work, we ask whether these results can be extended to heterogeneous graphs, which encode multiple types of relationship between different entities. We propose Neighbor Averaging over Relation Subgraphs (NARS), which trains a classifier on neighbor-averaged features for randomly-sampled subgraphs of the "metagraph" of relations. We describe optimizations to allow these sets of node features to be computed in a memory-efficient way, both at training and inference time. NARS achieves a new state of the art accuracy on several benchmark datasets, outperforming more expensive GNN-based methods