Design And Implementation Of User Social Group Constructing Method Based On Mobile Call Records

With the popularization of mobile terminals,the number of mobile users becomes saturated,which causes mobile carriers to conduct fierce competitions for developing new users,retaining old users,and finding new business supporting points,etc.However,mobile carriers have become the data transmission pipeline due to the rapid development of OTT(Over The Top).This brings new challenges to mobile carriers,because they cannot touch enormous valuable data transmitted in the pipeline.Such complex situations promote mobile carriers to focus on user experiences rather than businesses.It means that mobile carriers should understand users deeply and provide better services.For this,constructing user social group based on mobile call records attracts wide attentions of mobile carriers.Given the above,social group constructing method based on mobile call records is designed in this thesis.Firstly,this thesis preprocesses original call records,and makes records reflect user social wishes more clearly.In addition,taking full considerations of contact strength and contact periodicity,the corresponding model to measure contact tightness between users is designed through analyzing metrics including call duration,call times,and call type,etc.At last,each user is treated as a node and the contact tightness calculated before between users is regarded as the weight edge.Then a complex network based on social relations is abstracted.In order to construct the group based on weighted complex network and ensure the quality of group constructing results,referring to excellent community discovery algorithms in complex network,this thesis improves seed group constructing algorithm based on the principle of clique percolation,and combined with seed groups,group constructing algorithm based on the principle of local extension and group constructing algorithm based on the principle of label spread are improved and designed respectively.In addition,mobile call records are too many that any single high-performance computer might not process it effectively.So referring to Hadoop distribute system architecture,all methods proposed in this thesis are designed in parallel based on MapReduce programming model.In this thesis,social group constructing method based on mobile call records is implemented in parallel based on MapReduce,and benchmark simulated network,simulated call records and real call records obtained from China mobile of Liaoning branch are used for testing performances of social group constructing method designed in this thesis.Experiment results show that social group constructing method has good performances regardless of simulated or real data set.And it is feasible and effective.