Influence Maximization in Complex Networks Through Supervised Machine Learning

Abstract

Identifying influential nodes in complex networks is a well studied problem in network science. Finding an optimal set of influential nodes is an NP-Hard problem and thus requires the use of heuristics to find the minimal set of nodes capable of maximizing influence in a network. Once identified, these influencer nodes can been applied in various applications such as controlling disease outbreaks, identifying infectious nodes in computer networks, and finding super spreaders for viral marketing in social networks. This paper proposes a novel approach to solve this problem by modeling it as a supervised machine learning problem. Several synthetic and real world networks with nodal and network level attributes are used to train supervised learning models. Model performance is tested against real world networks emanating from a variety of different domains. Results show that the trained models are highly accurate in identifying influential nodes in networks previously not used for training and outperform commonly used techniques in the literature.