Research On IPTV User Behavior Based On Topic Model And Word2Vec

With the development of the Internet,Internet Protocol Television(IPTV)has rapidly popularized and attracts a large number of users for its massive video resources and high quality services.As network information carrying capacity grows rapidly,video resources of IPTV emerge ceaseless,not only causing troubles for users to approach programs,but also negatively affecting user's satisfaction with the IPTV.Therefore,it is necessary to study the viewing behavior of IPTV users,which can excavate the actual needs of users and improve their viewing experience.The study on the viewing behavior of IPTV users can mine their interest in different time periods,as well as cluster users by their viewing interest and analyze the characteristics of different groups.According to the characteristics of IPTV user viewing behavior data,this paper studies two aspects of IPTV user viewing behavior,one is to mine user behavior pattern and the other is to establish user interest model.This paper mainly carries on the following research:(1)This paper proposes a user behavior pattern mining method based on Time-Duration Coupled LDA.The Time-Duration Coupled LDA model add time and duration factors to the topic model LDA.In this model,time,duration and program record are jointly determined by the latent variables of user interesting and time periods,which are inferenced via Gibbs sampling.The superiority of proposed model on obtaining relevant interest topics and mining the user viewing interest distribution in different time periods is validated on IPTV user behavior dataset.Compared to the traditional c LDA,the proposed model is more effective on IPTV program recommendation.(2)This paper establishes a Multi-factor User Interest Model.The model uses Word2 Vec to mine the correlation between programs and quantifys the contribution of multiple viewing behaviors to user interests.First,use the Word2 Vec model to learn the vector representation of the IPTV program based on the contextual correlation between the record in the user viewing program records.Then,quantify the contribution of different viewing behaviors to user's interest,mainly including various behaviors in the viewing records(viewing,browsing,collecting,duration)and the frequency of the program appearing in the records.The paper proposes a quantitative strategy for duration by combining the user viewing duration and the duration of the program.Finally,generate IPTV user vectors by weighting IPTV program vectors based on TF-IDF method and multiple behavior characteristics,so as to establish Multi-factor User Interest Model.The experimental results show that the IPTV program vector can accurately characterize the similarity between programs.Users with the same age,gender,and interest can be clustered by Multi-factor user interest vector through clustering algorithm,and then the characteristics and viewing preferences of thier family members can be analyed.