Link Prediction Methods from Heuristics to GNNs: The Evolution of Link Prediction Methods - PhD Comprehensive Exam by: Nahid Abdolrahmanpour Holagh

PhD. Comprehensive Exam by: Nahid Abdolrahmanpour Holagh

Link prediction is a fundamental task in network analysis, aiming to forecast future or missing links between entities by analyzing the existing structure and dynamics of the network. This study will present an in-depth study of link prediction, tracing its development from early heuristic-based approaches to advanced machine learning and probabilistic models. Emphasis will be placed on recent progress in Graph Representation Learning (GRL), especially the emergence of Graph Neural Networks (GNNs), such as Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), and GraphSAGE. These models have significantly improved the ability to capture complex graph structures and node features for more accurate and scalable link predictions.

 

Furthermore, this study will focus on Dynamic Link Prediction (DLP), which extends traditional link prediction to time-evolving networks. We will explore how GNN-based methods are applied to DLP, considering their capability to model structural and temporal patterns. A comparative analysis will be provided, evaluating these methods in terms of scalability, handling of noisy and incomplete data, and their effectiveness in dynamic and heterogeneous networks.

Keywords: Link Prediction; Graph Neural Networks (GNNs); Dynamic Link Prediction (DLP); Dynamic Networks.