Research On IPTV Personalized Recommendation Method Based On User Interest Model

With the rapid development of the Internet,Interactive Personality TV(IPTV)has become the platform of preference for end users to watch videos.However,the vast amount of video resources on IPTV has also become an obstacle for end users to find videos of their real interest within a short time.Therefore,one of the difficulties for IPTV is to help end users identify target shows,and thus improve user experience with the assistance of information technology.Personalized Recommendation is the most effective way to solve this problem.It makes the accurate recommendation of video games possible by referring to end users' historical record,exploring users' interest and thus building a proper user interest model.This dissertation is dedicated to the personalized recommendation based on user Interest Model to improve the accuracy and efficiency of the personalized recommendation of IPTV.Firstly,pre-processing is conducted for user's view history to clear the useless data and thus improve data quality.Secondly,view images are adopted to represent users' view behaviors and categorization of images based on Convolutional Neural Network is also implemented to categorize IPTV users.Then,different interest models are established for users with and without stable view habits.Finally,historical view behaviors from IPTV users has validated the effectiveness of the method as proposed in this dissertation.Innovation of this dissertation is mainly reflected in below two aspects:1)Considering the large amount of IPTV users' historical view data,the complexity of family members' interests as well as the instability of view habits,this dissertation,for the first time,tries to categorize users' view behaviors by the adoption of view images.What's included in this view image is the timing,duration as well as contents of users' view history.Since the view image contains large amounts of user behavior information,and meanwhile,considers the fact that Convolutional Neural Network could realize the categorization of user view images through the function of convolutional layer and pooling layer,therefore,the methodology of Convolutional Neural Network has been adopted.Experimental results indicate that the accuracy of user categorization has been close to 90%.2)User interest model mining method based on rules of clustering and association has been proposed for users with stable view habits.Multiple factors such as view interests,timing and duration of users have been taken into full consideration so as to conduct comprehensive exploration into users' historical view behaviors based on associated information data,build accurate IPTV user interest model and thus improve the accuracy of recommendation of videos.Experimental results indicate an obvious improvement in the accuracy of recommendation of videos.