SFCLTA: Spectral Fusion Contrastive Learning with Topology-Adaptive Graph Augmentation

Graph Neural Networks (GNNs) have achieved remarkable success in graph analysis due to the Message-Passing (MP) mechanism, yet they struggle with heterophilic graphs where connected nodes often have distinct labels or dissimilar attributes. Graph Contrastive Learning (GCL) serves as a promising approach to extract the information beyond neighboring nodes, effectively mitigating the limitations of the MP mechanism in handling heterophilic graphs. Nevertheless, GCL faces two critical challenges when applied to heterophilic graphs, i.e., the potential distribution shift from data augmentation and the loss of robustness caused by high-frequency signals. To address these problems, we propose a novel model, namely the Spectral Fusion Contrastive Learning with Topology-Adaptive Graph Augmentation (SFCLTA) for unsupervised graph representation learning. Our method dynamically adjusts graph structures via a heterophily-aware augmentation strategy, and constrains high-frequency distortions by spectral regularization. We utilize the confidence-weighted fusion to enhance the robustness. Additionally, we introduce a feature reconstruction task as a prerequisite to explicitly mitigate feature-level distribution shifts. Experiments on multiple real-world datasets demonstrate that the proposed SFCLTA consistently outperforms baseline models in multiple tasks. Our code is released on https://github.com/JonathanGXu/SFCLTA.

Traditional graph neural networks usually struggle with heterophilic graphs where connected nodes have distinct attributes. Moreover, most existing graph contrastive learning methods tend to distort the critical high-frequency information during the augmentation process, leading to unstable training and degraded performance. To address these drawbacks, this paper proposes a novel SFCLTA model, which is an unsupervised learning framework. This model utilizes topology-adaptive augmentation to preserve high-frequency signals, applies confidence-weighted fusion to integrate multi-frequency features, and adopts feature reconstruction to align representations. This work enables more reliable graph analysis for real-world heterophilic data, supporting downstream tasks in social, biological and technological networks.