Effective Influence Maximization Based on the Combination of Multiple Selectors

Influence maximization is an extensively studied optimization problem aiming at finding the best k seed nodes in a network such that they can influence the maximum number of individuals. Traditional heuristic or shortest path based methods either cannot provide any performance guarantee or require huge amount of memory usage, making themselves ineffective in real world applications. In this paper, we propose MSIM: a multi-selector framework which combines the intelligence of different existing algorithms. Our framework consists of three layers: (i) the selector layer; (ii) the combiner layer, and (iii) the evaluator layer. The first layer contains different selectors and each selector can be arbitrary existing influence maximization algorithm. The second layer contains several combiners and combines the output of the first layer in different ways. The third layer evaluates the candidates elected by the second layer to find the best seed nodes in an iterative manner. Experimental results on five real world datasets show that our framework always effectively finds better seed nodes than other state-of-the-art algorithms. Our work provides a new perspective to the study of influence maximization.