SlotGAT: Slot-based Message Passing for Heterogeneous Graph Neural Network

TL;DR

SlotGAT introduces a novel slot-based message passing mechanism for heterogeneous graph neural networks, effectively preserving node-type semantics and improving performance on node classification and link prediction tasks.

Contribution

The paper proposes SlotGAT, which uses separate message passing slots for each node type and a slot attention mechanism, addressing semantic entanglement in heterogeneous GNNs.

Findings

Outperforms 13 baselines on 6 datasets.

Effectively preserves semantics across node types.

Enhances node classification and link prediction accuracy.

Abstract

Heterogeneous graphs are ubiquitous to model complex data. There are urgent needs on powerful heterogeneous graph neural networks to effectively support important applications. We identify a potential semantic mixing issue in existing message passing processes, where the representations of the neighbors of a node $v$ are forced to be transformed to the feature space of $v$ for aggregation, though the neighbors are in different types. That is, the semantics in different node types are entangled together into node $v$'s representation. To address the issue, we propose SlotGAT with separate message passing processes in slots, one for each node type, to maintain the representations in their own node-type feature spaces. Moreover, in a slot-based message passing layer, we design an attention mechanism for effective slot-wise message aggregation. Further, we develop a slot attention technique after the last layer of SlotGAT, to learn the importance of different slots in downstream tasks. Our analysis indicates that the slots in SlotGAT can preserve different semantics in various feature spaces. The superiority of SlotGAT is evaluated against 13 baselines on 6 datasets for node classification and link prediction. Our code is at https://github.com/scottjiao/SlotGAT_ICML23/.