Survey Reviews GNN-Based Link Prediction Techniques and Applications

Graph Neural Networks (GNNs) have become the dominant approach for link prediction, a task crucial for inferring missing connections and anticipating future relationships within data. Despite their widespread use, existing literature lacked a systematic review specifically focused on the underlying GNN architectures and their application across diverse graph structures. This new survey aims to fill that gap. The paper introduces a novel taxonomy, classifying recent progress based on both techniques and applications. From a technical standpoint, it delves into key GNN encoder architectures, including GCN-based, GAE-based, GAT-based, and GFormer-based methods, detailing their respective advantages and limitations. On the application front, the survey emphasizes practical use cases in knowledge graphs and recommendation systems, illustrating the real-world impact of GNN-based link prediction. Beyond current methods, the review also addresses the ongoing challenges in the field and proposes promising avenues for future research. This comprehensive overview serves as a valuable resource for researchers and practitioners seeking to understand and advance the state-of-the-art in GNN-based link prediction.