Topic-based Influence Maximization In Academic Network

With the development of Internet, social network analysis has attracted a growing number of researchers. Nowadays, influence maximization in social network has become popular, which attracts many researchers. Influence maximization aims at digging the most influential nodes in social network, which begins with selecting k initial nodes, then propagating the influence throughout the social networks, finally achieving the greatest number of active nodes. Influence maximization problem has a wide range of important application scenarios in marketing, product recommendation, rumor control and so on.To solve this problem, researchers have done several related work. For example, researchers have proposed classic influence propagation models. Some greedy algorithms are proposed to maximize the influence of the social network.However, the classic influence propagation models are based on the social network topology without taking into account relationship between nodes and the topic. The spread of influence are largely affected by these relationships.Based on the traditional influence propagation model, we proposed a method to calculate the topic propagation probability and three topic-based influence propagation models. Firstly, we establish metagraph for the relationship between nodes, information and words. Secondly, we obtain topic via tensor decomposition. Thirdly, we propose a method based on topic factor graph to calculate topic-based propagation probability. Finally, we have three topical influence propagation models by introducing topic-based propagation probability and gain the influential node which maximize the spread of influence in the academic network by the topical influence maximization algorithm.Compared with the previous influence propagation models, our improved influence propagation models take the topic into consideration. As we can drawn from the result of experiment, influence in the improved topic-based models have better results for simulation and spread in real academic network.