A community-based link prediction approach utilizing multiscale modularity and local patterns

Link prediction is a fundamental task in network science, aiming to identify missing or future connections based on observed structures. While conventional approaches often rely on global topological or embedding-based features, they tend to overlook the critical role of community organization and localized connectivity in driving link formation. To address these limitations, this study introduces the Community-driven Link Automaton (CLA), a novel framework that integrates multiscale community detection with local connectivity patterns. CLA employs a multiresolution variant of the Louvain algorithm to capture both macro- and micro-level community structures, thereby overcoming the resolution limits of traditional modularity optimization. It further enhances predictive power by incorporating Cannistraci-Hebb (CH) indices, which emphasize short-range connectivity, resulting in a robust dual-layered methodology. Comprehensive experiments conducted on 550 real-world networks across six domains: biological, economic, informational, social, technological, and transportation demonstrate that CLA consistently outperforms state-of-the-art baselines on multiple evaluation metrics, including Area Under the Precision-Recall Curve (AUC-PR) and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC). The results highlight the essential contribution of community structure to link formation and underscore CLA’s scalability, adaptability, and practical value in analyzing complex systems.