ERMNF: A novel multiplex network fusion method based on edge relevance

Recently, multiplex networks have been widely used to represent real-world complex systems. While they offer valuable insights into complex systems, their multi-layer structure poses significant challenges for network analysis tasks. Network fusion process has emerged as a powerful tool for addressing this issue; however, most existing methods are inappropriate for large-scale multiplex networks and ignore the inter-layer structure. To address this problem, we propose an edge relevance-based multiplex network fusion (ERMNF) model, which transforms the multiplex network into a monoplex network while preserving its essential structural properties. ERMNF operates in three main phases. First, it collects the links from all layers into a single-layer network. Next, ERMNF determines the relevance of each link in the binary aggregated network using two different edge relevance models based on the concept of shortest paths. Finally, it removes irrelevant links using an edge reduction (ER) model, while maintaining the set of nodes. We evaluated ERMNF on eight real-world multiplex networks, comparing it with five well-known fusion methods. Our experiments were two-fold. First, we assessed the fused network's ability to preserve the original network's topological properties. Second, we evaluated its performance on various network analysis tasks, including influence maximization, link prediction, and community detection.