Multi-Topic Influence Maximization In Large Scale Social Networks

Influence maximization has been widely studied in social network analysis. However, most existing works focus on global influentials while ignoring the fact that different topics may have different influentials. Even though a few works take topics into consideration, they neglect that the product or event needed to spread in the network refers to multi-topics instead of only one topic. At the same time, the users in the network also have different interests deciding the result of the influence propagation.In this paper, we study the influence maximization of social network by incorporating the multi-topic information into Independent Cascade(IC) model and propose a multi-topic sensitive diffusion model, the Multi-Topic Sensitive Independent Cascade model(MTSIC). Based on the MTSIC model, we can find the influential nodes which are closer to the reality. The topical Hyperlink Induced Topic Search(topical HITS) algorithm is used to calculate the influence and conformity of each node in the proposed diffusion models. As the product or event needed to spread may have location limitation, so the locations of influentials nodes are also considered. Considering the traditional influence maximization algorithm is not suitable for multi-topic situation, then we propose the Activation Nodes Similarity algorithm(ANS). Given that calculating the influence maximization in large scale network is very time-consuming, a solution based on Spark, namely Parallelization of Multi-Topic algorithm(PMT), is proposed to improve the efficiency.Considering the deficiency of traditional measurement to depict the importance of the multi-topic, we propose a novel measurement SIS to illustrate the effectiveness of the multitopic influence maximization algorithms. Extensive experiments on DBLP and Twitter data show the MTSIC model can exactly simulate the activation of the nodes in reality and ANS algorithm can find out the influential nodes which are more likely to propagate in real world. The efficiency of PMT algorithm is also proved. The result of thorough experiments is the best evidence of the effectiveness and efficiency of our model and algorithms.