Robust Learning on Noisy Graphs via Latent Space Constraints with External Knowledge

2025-07-07

AI Summary

Graph Neural Networks (GNNs) suffer substantial performance degradation on graphs with noisy edges. To address this, we propose KGLR (Knowledge-Guided Latent Regularization), a robust GNN framework leveraging external knowledge to enhance resilience against edge perturbations. KGLR employs a dual-encoder architecture to jointly model the original graph and a knowledge-constrained auxiliary graph; introduces a latent-space discrepancy penalty to explicitly suppress overfitting to spurious edges; and incorporates clean-link priors to regularize the embedding space—extended to heterogeneous graphs for improved interpretability. The method seamlessly integrates graph encoding, contrastive learning, and knowledge-driven regularization. Extensive experiments on multiple benchmark datasets demonstrate that KGLR achieves an average accuracy improvement of 5.2% under moderate noise levels, significantly outperforming state-of-the-art noise-robust GNNs. Furthermore, KGLR validates its efficacy and biological plausibility on a real-world protein–metabolite interaction network.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Natural Language Processing: Safety and Robustness

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Graph Neural Networks (GNNs) often struggle with noisy edges. We propose Latent Space Constrained Graph Neural Networks (LSC-GNN) to incorporate external "clean" links and guide embeddings of a noisy target graph. We train two encoders--one on the full graph (target plus external edges) and another on a regularization graph excluding the target's potentially noisy links--then penalize discrepancies between their latent representations. This constraint steers the model away from overfitting spurious edges. Experiments on benchmark datasets show LSC-GNN outperforms standard and noise-resilient GNNs in graphs subjected to moderate noise. We extend LSC-GNN to heterogeneous graphs and validate it on a small protein-metabolite network, where metabolite-protein interactions reduce noise in protein co-occurrence data. Our results highlight LSC-GNN's potential to boost predictive performance and interpretability in settings with noisy relational structures.

Problem

• Improves GNN performance on noisy graphs using external clean links
• Reduces overfitting to spurious edges via latent space constraints
• Enhances predictive accuracy in noisy protein-metabolite networks

Innovation

• Incorporates external clean links for noisy graphs
• Uses dual encoders with latent space constraints
• Extends to heterogeneous graphs like protein-metabolite networks