A Multi-seed Nodes Selection Strategy for Influence Maximization Based on Reinforcement Learning Algorithms

Abstract

Identifying influential individuals in the dissemination of information is an important topic in the study of social networks. Up to now, most of the previous works of Influence Maximization on social networks has been limited to selecting seeds based on a certain structural feature of the networks. These algorithms only consider a certain structural feature and cannot effectively select suitable seeds on social networks when the network has complex and changeable structures. Most of them only get good results on a special kind of networks. In order to find the most suitable nodes as the initial seed nodes in various social networks, we designed a new seeds selection algorithm which is based on reinforcement learning (IMQ). Our approach takes advantage of the characteristics of reinforcement learning’s agent that can continuously interact with the environment, this algorithm can be adapted to select the most suitable nodes as seed nodes on various social networks. It fully considers the influence of the network structure characteristics on the influence propagation process, so that this method can select the best nodes as seeds on social networks with different topologies. In order to demonstrate the superiority of the approach, we conducted comparative experiments on six real social networks. Experimental results show that IMQ can be applied to various structural social network, and has stronger universality than traditional methods.