Learning Conjoint Attentions for Graph Neural Nets

TL;DR

This paper introduces Conjoint Attentions for graph neural networks, integrating structural interventions and higher-order correlations to enhance node feature importance assessment and improve representation learning.

Contribution

The paper proposes Conjoint Attentions and Graph Conjoint Attention Networks (CATs), a novel method that combines multiple structural cues for improved GNN performance.

Findings

CATs outperform state-of-the-art baselines on benchmark datasets.

Conjoint Attentions effectively incorporate structural information.

Theoretical validation confirms CATs' discriminative capacity.

Abstract

In this paper, we present Conjoint Attentions (CAs), a class of novel learning-to-attend strategies for graph neural networks (GNNs). Besides considering the layer-wise node features propagated within the GNN, CAs can additionally incorporate various structural interventions, such as node cluster embedding, and higher-order structural correlations that can be learned outside of GNN, when computing attention scores. The node features that are regarded as significant by the conjoint criteria are therefore more likely to be propagated in the GNN. Given the novel Conjoint Attention strategies, we then propose Graph conjoint attention networks (CATs) that can learn representations embedded with significant latent features deemed by the Conjoint Attentions. Besides, we theoretically validate the discriminative capacity of CATs. CATs utilizing the proposed Conjoint Attention strategies have been extensively tested in well-established benchmarking datasets and comprehensively compared with state-of-the-art baselines. The obtained notable performance demonstrates the effectiveness of the proposed Conjoint Attentions.