Dual GNNs: Graph Neural Network Learning with Limited Supervision

TL;DR

This paper introduces a dual GNN framework that enhances semi-supervised node classification performance under limited supervision and noisy graph structures by combining regular and spectral clustering-based GNN modules.

Contribution

It proposes a novel dual GNN learning framework that jointly leverages regular and spectral clustering-based GNN modules for improved performance with scarce labels and noisy graphs.

Findings

Significantly outperforms baseline GNNs with limited labels

Robust against noisy and corrupted graph structures

Effective across various GNN baseline models

Abstract

Graph Neural Networks (GNNs) require a relatively large number of labeled nodes and a reliable/uncorrupted graph connectivity structure in order to obtain good performance on the semi-supervised node classification task. The performance of GNNs can degrade significantly as the number of labeled nodes decreases or the graph connectivity structure is corrupted by adversarial attacks or due to noises in data measurement /collection. Therefore, it is important to develop GNN models that are able to achieve good performance when there is limited supervision knowledge -- a few labeled nodes and noisy graph structures. In this paper, we propose a novel Dual GNN learning framework to address this challenge task. The proposed framework has two GNN based node prediction modules. The primary module uses the input graph structure to induce regular node embeddings and predictions with a regular GNN baseline, while the auxiliary module constructs a new graph structure through fine-grained spectral clusterings and learns new node embeddings and predictions. By integrating the two modules in a dual GNN learning framework, we perform joint learning in an end-to-end fashion. This general framework can be applied on many GNN baseline models. The experimental results validate that the proposed dual GNN framework can greatly outperform the GNN baseline methods when the labeled nodes are scarce and the graph connectivity structure is noisy.