Multi-dimensional Community Discovery And Influence Analysis Oriented On Mobile Data

With the popularity of mobile communication network,it has become the main way for people's communication.It has already accumulated large-scale user data,including calls,text messages and internet-surfing information,etc.These data reflect people's social relations,thus as an important data source for social network study.This paper extracted users' communication data nearly a month within a city,taking user as a node,users' call relations as an edge,thus building a social network for community discovery.This paper raises distributed representation method to represent each user by analyzing the characteristics of social network,and measure users' similarity through their vectorization,then discover coarse-grained communities through clustering.Within each community found,tensor decomposition is used for multi-dimensional fine-grained division,identifying the family group,friend group,working group and some interest groups.Within each fine-grained community in each dimension,PageRank is conducted for users' influence analysis and ranking.Specific work is as follows.(1)Raising distributed representation method for social community discovery,to divide users into coarse-grained communities.Traditional methods are faced with high complexity,data sparseness and unclear division,this article adopts distributed representation method to represent users by vectorizing them.As user relationship is distributed on each dimension of a vector,data sparseness problem can be solved to some extent,with high efficiency to deal with large-scale data.(2)On the basis of coarse-grained division,tensor decomposition is used for multi-dimensional fine-grained division.Multi-dimensional heterogeneous mobile data allow us to get multi-dimensional information of users,thus to divide them into various social groups,and tensor decomposition is used to predict missing values to solve the data sparseness problem,finally we can get fine-grained social communities in the reconstructed tensor of each dimension.(3)Within each multi-dimensional fine-grained community,PageRank algorithm is conducted for users' influence analysis and ranking.A web link's jump is analogy to a mobile user's call relationship,namely,the more a user is called,the bigger his influence is.Through the influence analysis,we can mine influential users for targeted marketing,which has important commercial significance.