Research On User Behavior Analysis Strategy Based On IPTV

With the rapid development and continuous improvement of modern Internet technology,TV services have entered the consumption era of big video consumption with multiple channels based on the Internet.Internet Protocol Television(IPTV)has became an important carrier and traffic portal for the home Internet,which has great marketing value.In the face of massive user video data,how to filter unconscious behavior data,quickly locate user preference elements,and carry out accurate push,need to be studied urgently.Through in-depth mining,analysis and research on home Internet traffic data,operators can gain insight into the viewing preferences of home users,and mastering the viewing behavior and traffic characteristics of IPTV users in the home Internet.Then,operators can optimize and rationally configure network resources and adjust the service method and service content of video services,to improve the quality of user's viewing and to provide users with efficient,fast and highquality personalized services.This paper proposes an IPTV user behavior classification strategy based on deep learning for the diversity and complexity of IPTV services.The deep learning method is used to model and analyze the IPTV set-top box data,and the user preference classification is obtained.According to the user's viewing preference,a collaborative filtering hybrid recommendation algorithm based on improved BiasSVD and cluster user nearest neighbor is proposed,which effectively alleviates the scalability problem of user scoring matrix,obtains more accurate user prediction score,and finally provides users with accurate and rapid targeted recommendation services,the innovations of this paper are as follows:a)Aiming at the problem of low recognition rate of the existing IPTV user behavior classification methods,a method of IPTV video user behavior analysis based on deep learning is proposed.The method deeply perceives the degree of user preference,constructs an ideal user-on-demand video activity classification model,and obtains higher classification accuracy.b)According to the user's viewing preference classification,an improved BiasSVD collaborative filtering recommendation algorithm is proposed to obtain the target user prediction score.At the same time,according to the prediction error brought by the change of user interest,an improved clustering user nearest neighbor collaborative filtering algorithm is proposed by introducing the interest heat factor to obtain the nearest neighbor prediction score of the target user.c)The predicted score of the improved BiasSVD model is adjusted according to the average difference between the nearest neighbor predicted score and the actual score of the target user,which can obtain a more accurate target user prediction score,and finally provide the user with accurate prediction recommendation.