Modeling influence diffusion in networks for community detection, resilience analysis and viral marketing

The past decades have seen a fast-growing and dynamic trend of network science and its applications. From the Internet to Facebook, from telecommunications to power grids, from protein interactions to paper citations, networks are everywhere and the network paradigm is pervasive. Network analysis and mining has become an important tool for scientific research and industrial applications to diverse domains. For example, finding communities within social networks enables us to identify groups of densely connected customers who may share similar interests and behaviors and thus generate more effective recommender systems; investigating the supply-network topological structure and growth model improves the resilience of supply networks against disruptions; and modeling influence diffusion in social networks provides insights into viral marketing strategies. However, none of these tasks is trivial. In fact, community detection, resilience analysis, and influence-diffusion modeling are all important challenges in complex networks. My PhD research contributes to these endeavors by exploring the implicit knowledge of connectivity and proximity encoded in the network graph topology. Our research originated from an attempt to find communities in networks. After carefully examining real-life communities and the features and limitations of a set of widely-used centrality measures, we develop a simple but powerful reachabilitybased influence-diffusion model. Based upon this model, we propose a new influence centrality and a novel shared-influence-neighbor (SIN) similarity. The former differentiates the comprehensive influence significance more precisely, and the latter gives rise to a refined vertex-pair closeness metric. Then we develop an influence-guided spherical K-means (IGSK) algorithm for community detection. Further, we propose two novel influence-guided label propagation (IGLP) algorithms for finding hierarchical communities in complex networks. Experiments on both real-life networks and synthetic benchmarks demonstrate superior performance of our algorithms in both