GNN-Based Link Prediction: New Taxonomy Categorizes AI Techniques for Inferring Missing Network Connections
A comprehensive ArXiv review systematically categorizes Graph Neural Network (GNN) techniques for link prediction, covering underlying architectures and diverse graph structures. The taxonomy aims to improve recommendation systems, fraud detection, and social network analysis by better inferring missing or future connections in data.
Researchers have published a comprehensive review on ArXiv that systematically examines Graph Neural Networks (GNNs) for link prediction — the task of inferring missing connections in data, such as predicting friendships in social networks or detecting fraudulent transactions. Unlike prior surveys, this paper specifically targets the underlying GNN architectures and the variety of graph structures encountered in real-world applications.
The review introduces a novel taxonomy that categorizes recent advancements in GNN-based link prediction techniques. This framework helps organize methods by the type of GNN architecture used and the nature of the graph (e.g., homogeneous, heterogeneous, dynamic, or temporal). The authors argue that existing reviews lack this dedicated GNN-centric perspective, making their work a critical resource for researchers and practitioners.
Link prediction is a fundamental problem in many AI systems. Improving it can directly enhance recommendation engines (suggesting friends, movies, or products), strengthen fraud detection by identifying unusual transaction patterns, and advance social network analysis by forecasting future connections. The paper also discusses open challenges, such as scalability to large graphs, handling noisy or incomplete data, and generalizing across different domains.
For full details, the paper is available on ArXiv at https://arxiv.org/abs/2607.16198.