Pseudo Contrastive Learning for Graph-based Semi-supervised Learning

TL;DR

This paper introduces Pseudo Contrastive Learning (PCL), a novel framework for semi-supervised GNNs that improves pseudo-label quality by focusing on negative pair identification and topological weighting, leading to better performance on real-world graphs.

Contribution

The paper proposes a new Pseudo Contrastive Learning framework that enhances pseudo-label reliability and incorporates topological information for improved GNN training.

Findings

PCL consistently outperforms other techniques on five real-world graph datasets.

Incorporating topological weights improves negative pair separation.

PCL enhances semi-supervised GNN performance across various models.

Abstract

Pseudo Labeling is a technique used to improve the performance of semi-supervised Graph Neural Networks (GNNs) by generating additional pseudo-labels based on confident predictions. However, the quality of generated pseudo-labels has been a longstanding concern due to the sensitivity of the classification objective with respect to the given labels. To avoid the untrustworthy classification supervision indicating ``a node belongs to a specific class,'' we favor the fault-tolerant contrasting supervision demonstrating ``two nodes do not belong to the same class.'' Thus, the problem of generating high-quality pseudo-labels is then transformed into a relaxed version, i.e., identifying reliable negative pairs. To achieve this, we propose a general framework for GNNs, termed Pseudo Contrastive Learning (PCL). It separates two nodes whose positive and negative pseudo-labels target the same class. To incorporate topological knowledge into learning, we devise a topologically weighted contrastive loss that spends more effort separating negative pairs with smaller topological distances. Experimentally, we apply PCL to various GNNs, which consistently outperform their counterparts using other popular general techniques on five real-world graphs.