Determining seeds with robust influential ability from multi-layer networks: A multi-factorial evolutionary approach

摘要

There has been a great stream of literature in studying the dynamics of information diffusion processes attached on networked systems. And the corresponding influential seeds selection task can be modeled as the influence maximization problem. Effective diffusion models and methods are developed to detect powerful seeds from both single- and multi-layer networks. Meanwhile, some recent studies indicate that the structural destruction also perturbs seeds’ spreading procedures, and robust seeds are expected in daily applications. Based on a performance measure for the robust influence maximization problem, the existing solution intends to determine a specific damage percentage first; guided by which, solutions are provided. However, the generality of such method is non-guaranteed, and the knowledge in search processes towards multiple scenarios is completely neglected. Focusing on these deficiencies, the multi-tasking optimization theory has been introduced into the seed determination task from multi-layer networks. Multiple optimization scenarios are considered parallelly and the synergy between these tasks is exploited in the search process. Combining informative knowledge from both genetic and fitness domains, a multi-factorial evolutionary algorithm containing problem-directed operators, named MFEA-RIMm, has been designed to solve the robust influence maximization problem on multi-layer networks. Structural characteristics in the inputted network are also emphasized in the local refinement process. Empirical analyzes demonstrate the notable performance of MFEA-RIMm over existing methods, and diversified results can be obtained simultaneously to cater to challenges in multiple information diffusive scenarios.