Link Prediction in Social Networks Using the HTOA

Social networks are complex and dynamic data structures whose analysis holds significant importance. Among various tasks, link prediction stands out as a fundamental aspect of social network analysis, focusing on identifying potential future relationships between nodes. Despite considerable advancements in this field, existing methods often face challenges related to computational efficiency and prediction accuracy. This study introduces a novel approach to link prediction in social networks by employing the Heat Transfer Optimization Algorithm (HTOA) for topological feature selection. HTOA, inspired by the physical phenomenon of heat transfer, identifies an optimal subset of topological features to feed XGBoost machine learning model. The primary advantage of the proposed method lies in its ability to substantially reduce feature dimensionality without compromising prediction accuracy. Experiments conducted on two real-world datasets, Facebook and CollegeMsg, demonstrate that the proposed method, can reduce the number of utilized features by up to 75% while maintaining a prediction accuracy (AUC) of 0.99. Furthermore, this dimensionality reduction results in a significant improvement in prediction time (up to 73.8%) compared to traditional methods. Comparisons with other optimization techniques, such as the Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), reveal that the proposed method outperforms them in selecting key features and achieving faster convergence. These characteristics make the proposed method a suitable choice for real-world applications, including recommender systems, large-scale social network analysis, and interaction pattern detection. Additionally, the analysis of selected features provides valuable insights into the underlying mechanisms of link formation in social networks.