An Multi-Objective Optimization Method Of Influence Maximization In Social Networks

With the increasing popularity of online social networks such as Weibo, WeChat, etc., disseminating corporate marketing information through social networks (such as the disseminate advertisements or distribute free coupons) has becoming a mainstream network marketing strategy. State-of-the-art researches on social marketing normally based on a single perspective of structural influence maximization, that is to use the random walk, linear threshold model or other methods based on network topology to find seed users who have the largest range of influence. However, the propagation effect of marketing information is actually caused by combined effects of multiple factors. So factors as seed users’structural influence, users’interest in marketing information, enterprise budget, etc. will have significant impact on seeds selection decisions. How to trade-off between these factors is important for corporate social marketing.Influence maximization based on network topology is the main idea of current social marketing, whose main issue is to choose the most influential seed users based on indicators like degree centrality, betweenness centrality and so on, so that the marketing information can spread over a wide range. However, the effect of marketing information dissemination in social marketing will also be subject to users’ preferences, since users have stronger preference will adopt information better. Furthermore, enterprises usually have budget constraint in marketing, and different seed users have different costs, those users with higher influence will have stronger propagating capacity, but will also have higher costs.In this paper, we build a multi-objective optimization model which consider structural influence, user interest and enterprise cost. In order to improve the optimization effect, we firstly use some classical algorithms to obtain influential nodes as candidate seed users. Then we use the improved linear threshold model to calculate structural influence, user interest and cost. After that, NSGA-II, a multi-objective optimization evolutionary algorithm, is used to obtain Pareto non-domination solutions. The experimental results on four real social networks show that our method can provide a more flexible policy recommendations for enterprises.