Reserch On User Indentification And User Profiling For Smart TV Based On Community Detection Algorithm

With the eye-catching development of the Internet,new television technologies such as smart TV,bi-directional Set-Top Box(STB)are appearing.Moreover,the studies of program recommendation based on STB and user profiling are beginning to gather momentum.Previous methods mainly provide recommendations to TV users on the basis of regarding one account as one person,which ignores the feature of TV shared by more than one person in most instances.Thus,making a distinction between users,who share the same account but have individual preferences,has a valuable contribution to the improvement of recommendation system.Again,the less interaction between TV and users makes personal information difficult to obtain or even not.In view of the existing problems and challenges in the field of TV program recommendation,this thesis proposes a new method to identify users in a single TV account and conduct user profile,aiming to improve the performance of the program recommendation system by refining the family composition of TV account.The main contents of this thesis are as follows:1)In this thesis,we innovatively proposed the concept of program network diagram with program as node and program similarity as edge weight.And based on this network model,community detect algorithm is used to conduct the search of program community,which is the premise to complete the identification of TV family members and user profile.2)In view of the shortage of explicit information and the simplicity of program type label in the historical viewing data of TV users,this thesis improves the performance of the user identification and user profile by collecting the information of specific micro-blog users related to program.Data cleaning and feature extraction of set-top box data and micro-blog data are conducted according to the characteristics of the program network model.3)Aiming at the difference of information meaning between Electronic Program Guide(EPG)data and microblog data,this thesis designs a simple calculation method of comprehensive similarity between two programs.And by changing the weight ratio of the two types of data,we can observe the influence of microblog data on the performance of program recommendation.4)Hereafter,we design three recommended methods for comparison experiments,all of which are based on the same item-based recommendation algorithm.Results show that the performance of recommendation applying the proposed method is more effectively than previous approaches.