Data Science and Analytics Thrust Seminar | Adaptive Non-submodular Influence Maximization on Social Networks

Emerging social medias are based on online social networks, which have improved the efficiency of information propagation and quickly become one of the most important works in the mobile Internet era. Thus, influence maximization has become the core issue in this field and can be applied to network monitoring, rumor control, social recommendation, and so on. The traditional algorithm design for influence maximization depends on the submodularity of influence function. However, in many real applications, the influence function is no longer submodular. According to recent development, the research evolves from submodular to non-submodular, from non-adaptive to adaptive, from offline to online, from traditional graph algorithm to modern machine learning. In this talk, we start from non-submodular optimization, taking about some feasible strategies to deal with it, and then extend it to the adaptive version. We mainly focus on our recent finding in this area and include some desirable direction in the future by the way, such as online learning.