Identifying and Correcting Label Noise for Robust GNNs via Influence Contradiction

TL;DR

ICGNN tackles label noise in graph-based semi-supervised learning by introducing Influence Contradiction Score (ICS), which leverages both graph diffusion and representation affinity to detect noisy labels. A Gaussian Mixture Model then differentiates clean and noisy labels, and a soft neighbor-aggregation mechanism corrects detected noise while pseudo-labeling extends supervision to unlabeled nodes. The approach is validated across multiple datasets and noise settings, demonstrating superior accuracy, robustness to varying noise and label rates, and favorable runtime characteristics. The combination of structure- and attribute-level analysis, together with corrective and semi-supervised strategies, provides a practical and scalable solution for robust graph learning in the presence of noisy annotations.

Abstract

Graph Neural Networks (GNNs) have shown remarkable capabilities in learning from graph-structured data with various applications such as social analysis and bioinformatics. However, the presence of label noise in real scenarios poses a significant challenge in learning robust GNNs, and their effectiveness can be severely impacted when dealing with noisy labels on graphs, often stemming from annotation errors or inconsistencies. To address this, in this paper we propose a novel approach called ICGNN that harnesses the structure information of the graph to effectively alleviate the challenges posed by noisy labels. Specifically, we first design a novel noise indicator that measures the influence contradiction score (ICS) based on the graph diffusion matrix to quantify the credibility of nodes with clean labels, such that nodes with higher ICS values are more likely to be detected as having noisy labels. Then we leverage the Gaussian mixture model to precisely detect whether the label of a node is noisy or not. Additionally, we develop a soft strategy to combine the predictions from neighboring nodes on the graph to correct the detected noisy labels. At last, pseudo-labeling for abundant unlabeled nodes is incorporated to provide auxiliary supervision signals and guide the model optimization. Experiments on benchmark datasets show the superiority of our proposed approach.

Figures (9)

• Figure 1: Illustration of the proposed framework ICGNN. Our model consists of two steps: (a) Noise detection by influence contradiction: the original graph and $K$NN-based representation affinity graph undergo graph diffusion from attribute and structure levels. From these two perspectives, the ICS values are calculated and fused, followed by a Gaussian mixture model for accurately identifying potential noisy labels; (b) Noise cleaning and pseudo-labeling by neighbor aggregation: robust noise cleaning strategy and pseudo-labeling technique integrate predictions from neighboring nodes to correct the detected noisy labels and provides additional supervision.
• Figure 2: Robustness analysis against different levels of label noises on DBLP and Pubmed.
• Figure 3: The comparison $w.r.t.$ different label rates on DBLP and Pubmed.
• Figure 4: Comparison $w.r.t.$ different $\alpha$ on Coauthor CS, Amazon Photo, DBLP, Pubmed datasets.
• Figure 5: Loss distribution (a and c) and confidence distribution (b and d) when training a GCN on Amazon Photo.