Betweenness centrality-based Adamic–Adar similarity for link prediction in complex networks
The task of Link Prediction (LP), which predicts the formation of future links in graphs, involves forecasting potential connections between nodes based on the existing network structure. LP plays a crucial role in domains such as social, biological, and communication networks, where it helps uncover the underlying structure and evolution of complex systems. The Adamic–Adar (AA) index is one of the most widely used similarity measures for LP because of its simplicity and effectiveness in capturing local structural patterns. However, purely local similarity measures fail to account for the global structural roles of nodes, particularly those that act as brokers connecting otherwise weakly connected regions of a network. In this study, we present the Betweenness Centrality-Based Adamic–Adar similarity measure (AAB), an endpoint-centric extension of the classical Adamic–Adar similarity measure for topology-based link prediction. The proposed method integrates the Adamic–Adar index with the betweenness centrality of candidate endpoint nodes, thereby modeling a dual mechanism of link formation driven by both local neighborhood reinforcement and global structural brokerage. Unlike existing centrality-enhanced approaches that primarily incorporate the centrality of common neighbors, AAB explicitly models the structural influence of the candidate endpoint nodes themselves. The proposed method is evaluated on multiple real-world networks from diverse domains using the Area Under the Receiver Operating Characteristic Curve (AUROC) and the Area Under the Precision–Recall Curve (AUPR). Experimental results demonstrate that AAB consistently outperforms the traditional Adamic–Adar index as well as several recent topology-based link prediction methods, highlighting the effectiveness of combining local similarity with global structural brokerage for accurate link prediction in complex networks.