Research On Influence Maximization Based On Target Optimization

We are currently in the information era,more and more social Apps are appearing in peoples 'lives.Different social relationships have built a huge social network for human beings.In social network,everyone has a different role and relationship with others.Social network relationships have shortened the distance of human beings which makes rapidly information propagating becomes possible.Therefore,how to disseminate important information through social networks has become the focus of research scholars.By simulating the strategy of marketing sale,scholars have raised the problem of Influence Maximization.The research goal of influence maximization is mainly to find some influential seed nodes in network.These nodes can maximize the influence of surrounding neighbor nodes under a specific propagation model.Therefore,most of the current works on influence maximization are centered on the topology of the network,and finding the seed nodes through some important index of social networks.Meanwhile,most of the current works are only finding some influential nodes and pay less attention to other goals.In fact,influence maximization problem is not a simple singleobjective optimization problem.Enterprises need to consider the cost factor when selecting influential users in marketing.How to find the influential nodes on the basis of minimum cost are more in line with realities.So,research based on multi-objective influence maximization problem also deserves our attentions.The practical significance of researching the influence maximization is very important.At present,the research results of influence maximization have been widely used in personalized recommendation,public opinion monitoring and recommendation systems and have achieved significant research results.Nowadays,with the continuous development of information technology,the social network relationships have become larger and larger,and the research of influence maximization has also put forward new requirements.Influence maximization problem is NP-hard problem which should be solved by Monte Carlo simulations to get the final influence spread.The intelligent optimization problem is outstanding in solving NP-hard problems.So,it is a new direction to solve influence maximization problem by using intelligent optimization problems.The current intelligent algorithms may exist some problems like more iterations or slow convergence speed or poor ability in optimization.To overcome the shortcomings of current works,we propose a new way to solve influence maximization problem.The main works of this paper is as follows.(1)In this paper,we first explain the concept of influence maximization problem and the judge index,then we introduce the propagation and classical algorithms in influence maximization.We conduct the shortcomings of algorithms in single-objective influence maximization and multi-objective influence maximization.(2)This paper proposes an algorithm based on clonal selection theory(CSAIM).In this algorithm,we first preprocess the datasets,using community detection algorithm to split and filter important communities.Then we calculate every node by eigenvector centrality to construct candidate seed node pool.Finally,we use clone selection algorithm to select seed nodes.Due to the high quality and small number of nodes in candidate node pool,the algorithm guarantees the influence spread and running time.Experimental results in three public datasets demonstrate the effectiveness of CSAIM.(3)This paper proposes an algorithm(PRNSGA-II)to solve the problem of influence maximization and cost minimization.In real life,the cost of finding nodes must be considered.How to select nodes with limited budget is the focus of our research.The PRNSGA-II first deletes the small communities and isolated nodes in datasets which reduces the size of network.Then we combine PageRank and EDV to introduce the first optimization function to calculate influence spread.According to the concept of degree centrality,we design a cost function as our second optimization function.After non-dominated sorting,crossover and mutation operations are performed to update the population.Iterative optimization is performed continuously to obtain the final Pareto Optimal Solutions.Experimental results in three public datasets shows the effectiveness of PRNSGA-II.