A hybrid dynamic memetic algorithm for the influence maximization problem in social networks

As an important research topic in social network analysis, the influence maximization problem is aimed at identifying a set of influential nodes from a social network that can produce the maximum propagation cascade effect. Developing flattering policies that can balance both the solution quality and computational efficiency is one of the key concerns on this problem. To address the deficiency of the classical static memetic algorithms that only consider the quality of neighboring nodes but lack the adaptability to different network topologies, this paper proposes a hybrid dynamic memetic algorithm (HDMA) for the influence maximization problem. According to the proposed framework, each node in the network is first assigned a label to indicate its importance. Based on the local influence, the potential influential nodes are added to the candidate set by ranking the importance, which is adopted to improve the quality of the initial population and prune the search space. An adaptive roulette taboo strategy by combining the stochastic local searching method, greedy mechanism with the centrality metrics is constructed to enhance the adaptability of the algorithm as well as its accuracy. The results prove that the HDMA can adapt well to different network structures. Higher accuracy solutions are obtained in faster running time, which verifies the satisfying performance of the algorithm in balancing global search and local exploitation.