Graph Augmentation for Recommendation

TL;DR

The GraphAug framework incorporates a graph information bottleneck (GIB)-regularized augmentation paradigm, which automatically distills informative self-supervision information and adaptively adjusts contrastive view generation, and consistently unveil its superiority over existing baseline methods.

Abstract

Graph augmentation with contrastive learning has gained significant attention in the field of recommendation systems due to its ability to learn expressive user representations, even when labeled data is limited. However, directly applying existing GCL models to real-world recommendation environments poses challenges. There are two primary issues to address. Firstly, the lack of consideration for data noise in contrastive learning can result in noisy self-supervised signals, leading to degraded performance. Secondly, many existing GCL approaches rely on graph neural network (GNN) architectures, which can suffer from over-smoothing problems due to non-adaptive message passing. To address these challenges, we propose a principled framework called GraphAug. This framework introduces a robust data augmentor that generates denoised self-supervised signals, enhancing recommender systems. The GraphAug framework incorporates a graph information bottleneck (GIB)-regularized augmentation paradigm, which automatically distills informative self-supervision information and adaptively adjusts contrastive view generation. Through rigorous experimentation on real-world datasets, we thoroughly assessed the performance of our novel GraphAug model. The outcomes consistently unveil its superiority over existing baseline methods. The source code for our model is publicly available at: https://github.com/HKUDS/GraphAug.

Figures (7)

• Figure 1: The overall architecture of the GraphAug framework. (i) The GIB-regularized graph augmentation is realized by integrating the learnable augmentor $\textbf{Aug}(\mathcal{G})$ and graph sampling reparameterization. (ii) Graph mixhop encoder enables mixing high-order relations for adaptive message passing. (iii) GIB-regularized contrastive optimization ($\mathcal{L}_\text{CL}$) optimizes GraphAug.
• Figure 2: Ablation study of sub-modules in GraphAug.
• Figure 3: The degradation in relative performance with respect to the noise ratio is examined. In this analysis, the topology of the user-item interaction graph is intentionally compromised by the introduction of randomly generated fake edges. The proportion of these fake edges is selected from the range of values: {0.05, 0.1, 0.15, 0.2, 0.25}.
• Figure 4: We examine the model convergence on the Gowalla.
• Figure 5: Hyperparameter study of GraphAug on Gowalla.