Enhancing Contrastive Link Prediction With Edge Balancing Augmentation

TL;DR

This paper introduces a theoretically grounded contrastive learning framework for link prediction that incorporates edge balancing augmentation to address node degree disparities, leading to improved performance on benchmark datasets.

Contribution

It provides the first formal theoretical analysis for contrastive link prediction and proposes a novel edge balancing augmentation method integrated into a new contrastive learning approach.

Findings

CoEBA outperforms state-of-the-art models on 8 benchmarks.

Theoretical analysis generalizes to autoencoder-based models.

Edge balancing improves contrastive learning effectiveness.

Abstract

Link prediction is one of the most fundamental tasks in graph mining, which motivates the recent studies of leveraging contrastive learning to enhance the performance. However, we observe two major weaknesses of these studies: i) the lack of theoretical analysis for contrastive learning on link prediction, and ii) inadequate consideration of node degrees in contrastive learning. To address the above weaknesses, we provide the first formal theoretical analysis for contrastive learning on link prediction, where our analysis results can generalize to the autoencoder-based link prediction models with contrastive learning. Motivated by our analysis results, we propose a new graph augmentation approach, Edge Balancing Augmentation (EBA), which adjusts the node degrees in the graph as the augmentation. We then propose a new approach, named Contrastive Link Prediction with Edge Balancing Augmentation (CoEBA), that integrates the proposed EBA and the proposed new contrastive losses to improve the model performance. We conduct experiments on 8 benchmark datasets. The results demonstrate that our proposed CoEBA significantly outperforms the other state-of-the-art link prediction models.