The Research On Influence In Online Social Network With Multiple Types Of Information

In recent years,various social services based on online social networks are changing our daily life.It also attracts researchers from computer science,sociology and eco-nomics to analyze and study online social networks.Different from the traditional media,users are producers,publishers and spreaders of information in online social networks,and the information propagation depends on users' social influence.There-fore,the influence analysis of online social networks is one of the key problems in social network analysis,and the observation and technique can be applied in many related applications,such as recommendation system,link prediction,community de-tection,viral marketing and emergency detection.The existing work on the influence analysis of online social networks usually col-lects data and does research only in one particular social network,while in the real situation,this assumption has obvious limitation.With the emergence of a large num-ber of online social networks,users will join in different social networks based on their different social needs.Therefore when we assess the social influence of an individ-ual,it is not enough to consider his influence in one single network,and information from other network is welcome to complete the users' profile.Similarly,in information propagation process,information will not stay in one social network,its propagation range should be extended to multiple networks.Therefore,in this paper,we combine multiple types of external information to study three main steps in influence analysis of online social network,including indi-vidual spreading influence,influence diffusion and individual accepting influence.We introduce real social network data such as Twitter and Foursquare as well as TV data to observe and analyze the influence in online social networks.The results of observation and analysis have led us to rethink and solve some important problems related to social network influence:key node mining,influence maximization,community detection and user social behavior prediction.Specifically,the innovations of this paper are as follows:1.Key Node Mining according to Cross-Network InfluenceIn the step of the individual spreading influence,the first work of this paper is measuring users' social influence and mining key nodes with high influence in the cross-network information propagation.These users who connect different networks and act as bridges are called "tipping users".In order to describe the information diffusion process in heterogeneous networks,we designed a cross-network information propagation model,which extracts various information dif-fusion channels in heterogeneous networks and learns the weights of various relationships from the data to calculate the users' activation probabilities.We also propose a new method to identify the tipping users who bring about the largest influence gain.Experiments on real social network dataset demonstrate the effectiveness of the model and the algorithm.2.Influence Maximization Problem in Information Cross-network DiffusionInfluence maximization problem is the classic application of research on infor-mation diffusion in social network.In real situation,viral marketing often meets the challenge that the target network structure is unknown and marketing cost is too high.Therefore,in this paper,we choose seed users from the target network and other external source network,and conduct viral marketing in a roundabout way in the target network.Information can not only propagate within the target network,but can also spread via the shared users with external networks,and activate users in the target network indirectly.We propose a intra-and inter-network information propagation model based on the random walk,and design an algorithm based on dynamic programming to solve the problem of maximiz-ing the cross-network influence.Experiments based on the real social network data prove the validity of the model and the algorithm.3.Community Detection based on Influence DiffusionCommunity detection is another application of influence diffusion.The third work of this paper studies the division of the emerging social network according to information interaction between users.As the structure of a new social net-work is sparse with little user information,it is difficult to measure the similarity between users.Therefore solving community discovery problem in emerging social network is very challenging.In this paper we propose to introduce use-ful information from certain mature networks to new networks to overcome the shortage of information.Taking full account of the differences between networks and based on an effective information diffusion model,we propose an efficient way to use information from other heterogeneous social networks to detect com-munities in the new networks.Experiments on real social networks show that our algorithm can effectively solve this problem.4.User Social Behavior Prediction with TV InformationIn the step of user accepting influence,we found that users in real life are often influenced by multiple sources of information.For example television adver-tising can inspire and affect the spread of information in social networks.The last work in this paper is to study how TV ads propagate in the online social network,how users discuss about TV ads in the network and to predict whether users will be infected with TV ads.In order to solve this problem,we propose a social influence inferring model for TV ads.The model first recognizes that the user's social activity is influenced by both online social friends and offline TV ads.And then the audience attitude towards advertising is divided into three different stages:cognitive,affective and conative.Based on the synergistic ef-fect of social friends and TV ads,the model uses a specific probability model to describe the audience attitude at each stage.Experiments on two real TV ads datasets demonstrate the effectiveness of the model.