Group-based Influence Maximization In Social Networks

Nowadays,online social networks have stepped into a golden age with booming development for Internet.People tend to share information over the social platforms such us Twitter,Facebook and Weibo,arousing much interest in the studies of information diffusion through the ”world-of-mouth” propagation.Among all the studies,influence maximization(IM)is the key information diffusion problem thanks to its commercial benefits.The goal of IM is to select k users as the seed set to maximize the influence spread for information(e.g.,advertisement),or the expected number of influenced users in social networks.On one hand,the studies on viral marketing can help companies choose influential promoters aiming at maximizing the expected spread for products,which turns into a well-known application for IM.On the other hand,the studies on rumor blocking and friend recommendation demonstrate the significance of IM on information diffusion research and social network development.Existing IM diffusion models rely on the edges between users as the social relationships to propagate the information.In other words,the probability that user u activates user v only depends on the ability of u to influence v.Nevertheless,people might not only follow the topics diffused by their friends but are also influenced in groups.One explanation is the conformity behaviors,where people conform to the opinions and actions of others by submitting to perceived group pressure.Conformity behaviors have been studied as parts of social psychology,affecting the information diffusion process from the users' perspective.There are previous works about conformity-based IM problem exploiting conformity characteristics to predict users' actions but ignoring user profiles.Therefore,we incorporate the user profiles into conformity-aware IM problems.Furthermore,conformity behaviors can be different types due to the reason why people conform.Based on the conformity-aware diffusion model,we propose a group-based influence maximization(GIM)algorithm to select k users as seeds who maximize the influence spread over the conformity-aware diffusion model.We integrate friend conformity and group conformity effects into our conformity-aware diffusion model based on user profiles and group profiling.In GIM algorithm,we rank the groups by their sizes,searching for the seeds with highest estimated influence spread.Empirically,our experimental results verify the effectiveness and efficiency for our GIM algorithm compared with other baseline IM algorithms at influence spread,attribute matching rate and running time on two synthetic datasets and one real world social network.In the future,we will consider the community detection and community search methods to learn the graph structure.