User Behavior Modeling And Programs Personalized Recommendation On IPTV Based On Tensor Decomposition

With the mutual penetration and gradual integration of telecommunication networks,cable television networks and computer networks,the Internet Protocol Television(IPTV)emerged.The accurate recommendation of IPTV programs guarantees the quality of experience(Qo E)for users.Facing a large number of IPTV programs,how to quickly locate the interest to avoid information overload issues is urgently researched.Traditional programs recommendation methods cannot be applied to IPTV programs recommendation.The reason is that the traditional recommendation methods are mainly to study the relationship between users and programs,without considering the user's viewing behavior at different times.Watching IPTV is a family behavior,that is,the members watching TV at different times may be different,and the interests of individual users are different at different times.At the same time,it is necessary to solve the problem of sparse viewing data of users.Therefore,in order to achieve personalized recommendation of IPTV programs,it is necessary to analyze the viewing behavior of IPTV users and explore the potential relationship among users,IPTV programs and viewing periods.This thesis takes the personalized recommendation of IPTV programs as its background.Based on the multiple challenges faced by IPTV programs recommendation,it has systematically studied the personalized recommendation of IPTV programs.The main work is as follows:(1)In order to achieve personalized and accurate recommendation of IPTV programs,we proposed a user's viewing preference model that combined with viewing duration,viewing period and service type.Based on the user's historical viewing log,we analyzed the intrinsic relationship among the IPTV program genres,the duration of the programs,the user's viewing time periods and the viewing duration.We deeply perceived the user's loyalty,interest and preference,and built the user's viewing preference model.(2)In order to improve the viewing experience of IPTV users,we proposed a personalized recommendation method TCIF that takes both time context and implicit feedback into account.Based on Tucker decomposition model,we combined time context with the user's viewing preference model to mine potential relationships among users,TV program genres and viewing periods,predicted the user's preference for various programs,and realized personalized recommendation for IPTV users.(3)In order to describe the influence of the group on IPTV users,we proposed a programs recommendation method RCGP that integrates regional characteristics and group preferences.We divided the user's address into four categories: urban,county,township and village,then combined with the user's viewing preference model to cluster the users.Extra constraints are embedded in the tensor decomposition model to reduce the sparseness of the tensor.Finally,we adopted the hierarchical tensor decomposition that takes the user's viewing preference and the group's viewing preference into account,so as to achieve personalized recommendation of IPTV programs.In addition,the research on user's behavior model and recommendation methods and the evaluation of performance are based on real user viewing data sets.Comparing TCIF proposed in the thesis with the existing algorithms,the results show that our accuracy is better than that of other algorithms by about 10%,while maintaining high coverage,diversity and novelty.Comparing the two proposed methods TCIF and RCGP,the results show that the accuracy of RCGP is at least 13% higher than that of TCIF,and saves a lot of memory space.