User Behavior And Service Correlation Analysis Based On Big Data

With the rapid development of communication technology, the human has entered the era of mobile Internet, most people use mobile terminals to access the mobile Internet for information acquisition and interaction. The huge mobile user group, not only determines the huge potential of the mobile Internet market,but also greatly promoted the research on the mobile Internet-related field. In addition, the combination of mobile Internet and big data create great value for operators. The use of big data analysis and processing technology, not only can optimize the network configuration and improve network performance, but also can provide users with personalized service to enhance user satisfaction.Therefore, in the context of big data, the analysis of user behavior and service correlation is of great significance.In this thesis, we study the service use behavior of mobile Internet users and the correlation of mobile Internet services from the real data of a large city in China. The main works are as follows:First, we establish a model that can characterize the service use behavior of mobile Internet users, and the results show that users tend to use the same service type and switch back and forth between familiar services in a short time.Second, combing the user behavior model with the Hidden-Semi Markov model (HSMM), we predict the behavior of mobile Internet users, and the results show that more than 80% of the users whose service use behavior has a prediction accuracy of 0.8 or higher for the selected 1500 users.Third, we analyze the variation of traffic volume in one day, divide a day into 103 time windows and establish the inter-service state transition model. The service occurrence probability matrix and the service state transition probability matrix are used to analyze the inter-service correlation. The results show that Web pages, instant messaging and streaming media services which occupy a large proportion of the services used, share a large correlation and that they also share a strong correlation with those which occupy a small proportion of the services used.Finally, through the way of Kalman filtering algorithm, we do the service occurrence probability prediction to verify whether the state transition matrixes can be a standard to describe the interactions between mobile services. The results show that the prediction average relative error is rather small, the magnitude of which is 10e-4, and this proves that the state transition matrixes can well reflect the correlation among those services.The research results of this thesis are helpful to improve the suitableness between operator's recommendation and user's preference, to provide more accurate policy support to improve the user satisfaction, to further guide the service provider to establish network evaluation and optimization index system strongly related with user experience.