Social Network Mining And Analysis Based On Telecommunication Data

In telecom network(Telco Network),the call detail records could be used to construct a pretty traditional social network.Given the comparative large-scale and social property of Telco data,it is natural to ask whether the social correlation theory could be applied to the telco social networks.If social correlation did exist in telco social network,providing a series of recommendation activities to telecom users is significant,which also gives new ways to telecom operators form another point of view.Aiming at the question of whether social correlations exist in telco social network,classic hypothesis testing method in statistics is used to authenticate the problem.Using t test and U test in one-tailed mode,it is verified that the similarity between two users who know each other is higher than two strangers in both traditional similarity methods and the hierarchical similarity method.Based on the conclusion that social correlations exist in telco social networks,a new social relation model integrated with user call mode factor is proposed in this thesis,which aims at the ignorance of call mode information in the past models.Through the social based recommendation system,the proposed social relation model achieves a better performance in user relationship description and Top-K recommendation task.Facing with the problem that traditional recommendation algorithms on a single computer cannot quickly complete recommend tasks with large-scale data,the ALS-SoRec algorithm is implemented on the distributed system.Via the prediction of interest preference of telco users,recommendation model training and assessment tasks for million users can be completed within an hour.Through the three parts work above,the existence of social correlation theory in Telco Network is confirmed from multiple perspectives.At the same time Internet preference for telco users is recommended with telco social network.Based on this series of work,telecom operators can push related content to telecom users according to the results of the recommend task.