Research On Social Influence Analysis And Its Applications In Social Networks

With the rapid development of the mobile Internet and mobile terminals,various types of online social networks have rapidly spread.The emergence of these social networks has greatly enriched the social needs of users.Unlike traditional information media,people can not only access information,but also publish and spread information in social networks.In the Internet age,online social networks have become important media for people to access information,spread information,and communicate with each other.With the widespread popularity of online social networks,the scale of social users has exploded,and a wealth of social activity data which contains important information has been generated.In-depth study of such data can be a good way to mine social influences of users in social networks,which will push the study of social applications,such as information diffusion,online marketing,community detection,behavior predictions of users,social recommendations and social advertising.Undoubtedly,these social applications would be of great help to social network platforms and social users.As social influence has great commercial value and application value,the research on social influence analysis and its application in online social networks has attracted extensive attention.Different types of online social network sites and diverse social data provide opportunities for research on social influence analysis and its applications in online social networks.At the same time,it also brings serious challenges.Although researchers at home and abroad have made a series of research results in this field,the existing researches still have shortcomings in the accuracy of social influence prediction and the depth of mining big data in social networks.Therefore,in order to meet the diversified application requirements and overcome the shortcomings of the existing researches,this research study the measurement of social influence in online social networks and two important social services,i.e.,viral marketing and social influence based community detection,by utilizing the topology structure of online social networks and the behaviors of social users.The main contributions of this research are listed as follows:(1)For the social influence measurement problem in online social networks,based on the unique features of event-based social networks,this research proposes a hybrid collaborative filtering based social influence prediction method.Firstly,in order to study the social in'fluence with fine granularity,a user's event-level social influence is defined as the proportion of the user's friends who are influenced by him to attend the event.Based on users'history behaviors,a user-event social influence matrix is constructed,in which each entry represents the influence value of a user on an event.Secondly,the social influence meausrement problem is formulated as the prediction of unobserved entries of the constructed user-event social influence matrix.In order to solve this prediction problem,a hybrid collaborative filtering prediction model is proposed,which incorporates both event-based and user-based neighborhood methods into matrix factorization.Thirdly,in order to further improve the prediction accuracy,an additional information based neighborhood discovery(AID)method by considering both event-specific features and user-specific features is proposed.In particular,the AID method utilizes multiple features of social events,i.e.,event content,event physical location and event organizer.Finally,the experiments based on the real-world dataset crawled in the famous event-based social network,i.e.,DoubanEvent,are carried out.The experimental results demonstrate that the proposed hybrid collaborative filtering model is effective in measuring the event-level social influences of users.(2)For the problem of influence maximization problem in online social networks,incorporating the preferences on topics and preferences on physical locations of social users,this research proposes a location-aware targeted influence maximization method.Firstly,for a given online query which contains a set of topics and a physical region,a TR-tree index structure is proposed which incorporates the topic information into the R-tree index structure to find the targeted users and compute the preferences of these users efficiently.Secondly,based on the location-aware targeted query and the computed preferences of users,an influence spread function for a seed set is defined.Thirdly,in order to solve the location-aware targeted influence maximization problem,two efficient approximation algorithms are proposed.In particular,based on the TR-tree index structure,the offline social influence indices,and the fast online social influence estimation method,the proposed approximation algorithms could find the seed set efficiently.Fourthly,in order to further improve the efficiency,a fast heuristic algorithm is proposed by estimating users' social influences in the whole social network based on their local social influences.Finally,experimental results conducted on real-world dataset demonstrate the effectiveness and efficiency of the proposed approximation algorithms and the heuristic algorithm.(3)For the community detection problem in online social networks,by considering the heterogeneous features of event-based social networks,this research proposes a social influence based community detection method.Firstly,based on the online social network and offline users' events participation network in event-based social networks,a social influence computation method is proposed which incorporates both of the social network topological structure and offline behaviors of users.In particular,based on the online social relations of event-based social networks,a network structure based social influence computation method is proposed;Based on the historical behaviors of users,a behavior based social influence computation method is proposed which considers users'preferences on three aspects,i.e.,topics of events,regions of events and organizers of events.Secondly,based on the computed social influence,a neighborhood constraint based deep auto-encoder method is proposed to obtain the community-oriented latent dimensional representations of users.Thirdly,based on the latent representations of users,the communities are detected by adopting the k-means clustering algorithm.Finally,experimental results conducted on real-world dataset demonstrate that the proposed algorithm could detect communities with dense connections of event-based social networks.