Research On Analysis And Prediction Algorithms Of User Service Behaviors For Mobile Internet Environment

In Mobile Internet Environment(MIE),each user can use a mobile terminal to rapidly access an Internet service through wireless communication technologies.With the increase of user demands,the service types become more and more diverse.However,different types of network services have completely different requirements of QoS metrics and network resources.Hence,the limited network resources cannot satisfy the QoS requirements of various services at once.Based on the analysis and prediction of complex and dynamic user behaviors,each user's next service type can be mined and then the optimal network resources can be reserved and allocated accordingly,so as to solve the above problem effectively.Therefore,this thesis investigates the analysis and prediction algorithms of user behaviors for MIE.The main works are given as follows.I?The basic contents of MIE are introduced briefly,and then the current researches of user behavior analysis and prediction are surveyed respectively.Furthermore,the related theoretical bases of this thesis are introduced detailedly.II?An improved fuzzy clustering theory based user behavior analysis algorithm is proposed.First,service interest similarity and service sequence similarity are respectively defined,and then integrated similarity metric is established.Second,a integrated similarity metric based fuzzy clustering model is formulated,which determines the initial cluster centers by using the grid partition method and adjusts the user cluster number according to the average user membership,so as to obtain user clusters rapidly and accurately.Simulation results verify the effectiveness of proposed algorithm.III?An improved Markov fusion model based user behavior prediction algorithm is proposed.First,a multi-order Markov prediction model for a single user is established,and then service preference degree is introduced to increase the prediction accuracy of the above model.Second,based on the integrated similarity metric,the nearest neighbor set of a target user is obtained,and then a multi-user and multi-order Markov fusion prediction model is formulated,so as to achieve the accurate predictions of target user's behaviors.Simulation results verify the effectiveness of proposed algorithm.Finally,the conclusions are summarized,and the future research directions are given.