Dynamic Network Influence Maximization Detection Algorithm Based On Regular Area Scale

Today's society has gradually entered the era of data,and data mining technology plays an irreplaceable role in business,economy,government and other fields.As an important branch of data mining,influence maximization has important applications in marketing,disease prevention,public opinion guidance and so on.The problem of influence maximization is to select k nodes in the network under the given budget condition K to maximize the influence expansion of the K nodes.With the in-depth study of this problem by scholars,although many effective methods have been put forward,most of them are based on static network,not suitable for the growing dynamic social network in reality,which leads to the problem has not been effectively solved.This master thesis focuses on the problem of influence maximization in dynamic networks:(1)An observation network structure is proposed to simulate the evolution of real network.The observation network with a time stamp represents the real network at the corresponding time.The local nodes in the real network are probed to update the observation network,so that the solution with the maximum influence on the observation network is close to the solution in the real network.(2)This master thesis proposes a RAS-Max G algorithm based on Regular Area Scale,which improves the effectiveness of obtaining node influence.In this master thesis,the weight sum of multi neighborhood nodes is used to calculate the influence value of nodes.The neighborhood betweenness of the node is obtained according to the Regular Area Scale value,and the influence value of the target node is obtained by adding the neighborhood weights of different levels.(3)The performance of the probed algorithm is optimized and the quality of the seed set is improved.The model first constructs the observation network to restore and simulate the real network,then calculates the estimated degree change value of each node based on the influence value of the nodes,finds the node with the greatest influence to enhance the seed set,probe and updates the observation network,and finally obtains the seed set by applying the influence algorithm on the observation network,so that the influence of the seed set spreads most widely in the real network.In this master thesis,large number of experiments are carried out on three real network datasets to verify the robustness of the improved algorithm: under the premise of the same size of seed set,the performance of the improved algorithm on different number of probe nodes is 5%-10% higher than that of the baseline algorithm;under the premise of the same number of probe nodes in each round,the performance of the improved algorithm on different size of seed set is better than that of the baseline algorithm.