LocalGCL: Local-aware Contrastive Learning for Graphs
Haojun Jiang, Jiawei Sun, Jie Li, Chentao Wu
TL;DR
LocalGCL tackles the tendency of graph contrastive learning to overemphasize global patterns by introducing a masking-based objective that highlights local graph information. It integrates a contrastive loss with a masking-based reconstruction objective in a multi-task framework using a shared GNN encoder, and it dynamically tunes the loss balance with a schedule that shifts from global to local emphasis during training. The method demonstrates strong performance on unsupervised graph classification and transfer learning tasks, outperforming several baselines and validating the importance of incorporating local structure in graph representations. This approach offers a practical and scalable way to obtain more comprehensive graph embeddings for downstream tasks across diverse domains.
Abstract
Graph representation learning (GRL) makes considerable progress recently, which encodes graphs with topological structures into low-dimensional embeddings. Meanwhile, the time-consuming and costly process of annotating graph labels manually prompts the growth of self-supervised learning (SSL) techniques. As a dominant approach of SSL, Contrastive learning (CL) learns discriminative representations by differentiating between positive and negative samples. However, when applied to graph data, it overemphasizes global patterns while neglecting local structures. To tackle the above issue, we propose \underline{Local}-aware \underline{G}raph \underline{C}ontrastive \underline{L}earning (\textbf{\methnametrim}), a self-supervised learning framework that supplementarily captures local graph information with masking-based modeling compared with vanilla contrastive learning. Extensive experiments validate the superiority of \methname against state-of-the-art methods, demonstrating its promise as a comprehensive graph representation learner.