Do Transformers Really Perform Bad for Graph Representation?

TL;DR

This paper introduces Graphormer, a Transformer-based model for graph representation learning that effectively encodes structural information, achieving state-of-the-art results and unifying many GNN variants.

Contribution

The paper presents Graphormer with novel structural encoding methods, demonstrating its superior performance and theoretical expressive power compared to existing GNNs.

Findings

Graphormer attains excellent results on large-scale graph benchmarks.

Structural encoding methods significantly improve Transformer performance on graphs.

Many GNN variants are special cases of Graphormer with appropriate encodings.

Abstract

The Transformer architecture has become a dominant choice in many domains, such as natural language processing and computer vision. Yet, it has not achieved competitive performance on popular leaderboards of graph-level prediction compared to mainstream GNN variants. Therefore, it remains a mystery how Transformers could perform well for graph representation learning. In this paper, we solve this mystery by presenting Graphormer, which is built upon the standard Transformer architecture, and could attain excellent results on a broad range of graph representation learning tasks, especially on the recent OGB Large-Scale Challenge. Our key insight to utilizing Transformer in the graph is the necessity of effectively encoding the structural information of a graph into the model. To this end, we propose several simple yet effective structural encoding methods to help Graphormer better model graph-structured data. Besides, we mathematically characterize the expressive power of Graphormer and exhibit that with our ways of encoding the structural information of graphs, many popular GNN variants could be covered as the special cases of Graphormer.