Learning-based Link Prediction Methods Integrating Network Topological Features and Embedding Representations

Abstract: Link prediction, as a frontier task in complex network topology analysis, aims to infer the existence of latent links between node pairs based on observed nodes and structural information. We propose an ensemble link prediction model that integrates network topology features and embedding representations (TELP), designed to overcome the limitations of conventional heuristic methods in capturing node attributes and deep structural patterns, as well as the weak interpretability and limited generalization of learning-based approaches. TELP leverages a multi-stage architecture. Local connectivity patterns are captured through network-type-aware selection of homogeneous and heterogeneous topology features, which also promotes interpretability. To incorporate global structure, Node2Vec embeddings are generated and fused with these topology features, resulting in comprehensive multi-dimensional representations. Building on this enriched feature space, an ensemble of logistic regression, random forest, and XGBoost models is deployed to maximize predictive performance and robustness. Experiments on nine classical benchmark networks demonstrate that TELP achieves superior AUC and AP performance compared with traditional heuristic approaches and mainstream graph neural network models, while ablation studies further confirm that feature fusion and ensemble strategies are essential for optimal performance.