Enhancing Network Role Modeling: Introducing Attributed Multiplex Structural Role Embedding for Complex Networks

Lili Wang, Chenghan Huang, Ruiye Yao, Chongyang Gao, Weicheng Ma, Soroush Vosoughi

Abstract: Numerous studies have focused on defining node roles within networks, producing network embeddings that maintain the structural role proximity of nodes. Yet, these approaches often fall short when applied to complex real-world networks, such as Twitter, where nodes have varying types of relationships (e.g., following, retweeting, replying) and possess relevant attributes impacting their network role (e.g., user profiles). To address these limitations, this study presents a novel method for attributed (for dealing with attributed nodes) multiplex (for dealing with networks with different types of edges) structural role embedding. This approach uses an autoencoder mechanism to concurrently encode node structure, relationships, and attributes, thus successfully modeling nodes’ roles within networks. Our method’s effectiveness is shown through quantitative and qualitative analyses conducted on synthetic networks, outperforming established benchmarks in identifying node roles within multiplex and attributed networks. Additionally, we have assembled a robust real-world multiplex network composed of almost all verified Twitter users comprised of retweet, reply, and followership interactions between these users, representing three different layers in our multiplex network. This network serves as a practical environment to evaluate our method’s capability to map the structural roles of users within real-world attributed multiplex networks. Using a verified dataset of influential users as a reference, we show our method excels over the existing benchmarks in learning structural roles on large-scale, real-world attributed multiplex networks, exemplified by our Twitter network.