Enhancing Link Prediction with Fuzzy Graph Attention Networks and Dynamic Negative Sampling

TL;DR

This work tackles suboptimal link prediction caused by random negative sampling in GNNs. It introduces FNS, a fuzzy rough-sets-based negative sampling strategy, and FGAT, a graph attention network that incorporates fuzzy set principles for robust neighbor aggregation. FNS computes a quality score for candidate negatives using fuzzy lower approximations and kernel-based similarities, with $Score(x,y)=\alpha\underline{R_B}d_y(x)+(1-\alpha)\underline{R_B}d_x(y)$ guiding selection. Experiments on Ca-netscience and Ca-sandi-auths show FGAT consistently outperforms baselines, with average metric gains of 7.11% and 15.55% respectively. The results validate the usefulness of fuzzy rough sets for dynamic negative sampling and representation learning in link prediction, suggesting broader applications in graph-based learning.

Abstract

Link prediction is crucial for understanding complex networks but traditional Graph Neural Networks (GNNs) often rely on random negative sampling, leading to suboptimal performance. This paper introduces Fuzzy Graph Attention Networks (FGAT), a novel approach integrating fuzzy rough sets for dynamic negative sampling and enhanced node feature aggregation. Fuzzy Negative Sampling (FNS) systematically selects high-quality negative edges based on fuzzy similarities, improving training efficiency. FGAT layer incorporates fuzzy rough set principles, enabling robust and discriminative node representations. Experiments on two research collaboration networks demonstrate FGAT's superior link prediction accuracy, outperforming state-of-the-art baselines by leveraging the power of fuzzy rough sets for effective negative sampling and node feature learning.

Figures (1)

• Figure 1: The FGAT Framework