Research On Influence Optimization And Prediction Of Social Networks

In recent years,social networks have become important platforms for people to make friends and cooperate because of their rich content and immediate information transfer with the development of the Internet.In addition,lots of academic researches about social networks have been launched to explore valuable information.One of the most important research topics in these academic studies is the influence analysis.For example,influence optimization and prediction analysis of social networks can help us to find influential users and understand how to diffuse information in a better way,which can be applied to solve a lot of hot issues such as expert discovery,product marketing,further improve the ability of utilization or management of social networks.Hence,based on the influence optimization and prediction research in social networks,this paper considers two valuable problems: individual-based influence prediction and group-based influence optimization.The contribution of our work is as follows:First,in terms of individual-based influence prediction,we select the typical representative of social network,i.e.academic social network.Considering that academic influence is of great significance in practice,and the present measure and prediction methods of academic influence are still not mature,we choose a fairer metric named g-index to measure individual academic influence.In the meanwhile,this paper proposes a novel problem that explores how to predict the future g-index with a given time interval.In order to solve this problem,we first extract different scientific performance measures from the academic social network.Secondly,a deeplearning based prediction model is designed to solve the problem by utilizing these extracted measures.Finally,the experiment results on real-world academic social networks show the considered problem is solvable.It is illustrated that a scholar's future g-index is related to the performance measures extracted from the current academic social networks.Specifically,future g-index has a positive correlation with the scholar's current g-index,the square root of the paper citation,the number of papers,and a negative correlation with the local clustering coefficient.Other factors such as the research years,the average citations per paper and the average number of cooperation have little correlation with the future g-index.In addition,it is verified that the proposed deep learning based prediction model achieve a better performance than those classic machine learning models.Second,in terms of group-based influence optimization.Considering that the existing influence optimization problem can not meet the requirements of the practical product marketing,this paper proposes a novel influence optimization problem named dominated competitive influence maximization problem.The considered problem researches how to find k-size seed set of target product so as to maximize the difference between the influence of desired product and its competitors.In order to solve it,a new propagation model is proposed,which considers time limit,time delay and proportional probability activation principle.Under the proposed propagation model,the considered problem is a NP-hard problem.Fortunately,it is proved that the objective is monotonic and submodular,which makes it possible to solve the considered problem by designing a greedy algorithm with at least 1-1/e approximation precision.Furthermore,in order to reduce the complexity of greedy algorithm,an algorithm is put forward to solve the considered problem.Finally,extensive experiments are conducted on four real-world datasets,proving that the considered problem is solvable.In addition,simulation results demonstrate that the proposed algorithm achieves better performance than high degree algorithm and random algorithm.