Research On The Recommender System Of Live TV

Live TV is a popular activity for people,with a wide range of users around the world.However,with increasing of live TV channels,the phenomenon of "information overload" in TV shows is prominent,and it is difficult for audiences to choose their favorite TV shows.As an effective way to solve the problem of "information overload",recommender systems have been widely studied and applied in many fields.This paper focuses on the research on the recommendation of live TV programs,aiming to recommend the favorite programs to the viewers who are switching channels.By simulating the viewing scene of live TV,considering the cold start of TV show and the sparse viewing data,this paper focuses on the measurement of implicit feedback of live TV,multi audience shared account,and the utilization of positive and negative feedback of viewing behavior.First of all,for the measurement of implicit feedback in live TV,the existing methods do not take into account the redundant time at the end of the video(such as advertising time),resulting in significant deviation in the measurement results.Therefore,a collective decision based measurement method of implicit feedback in live TV show is proposed.By using the collective decision of the audience,the end time of TV show is modified,so as to optimize the existing different types of implicit feedback measurement methods.Secondly,the multi-viewer sharing account problem is rarely considered by the existing methods of recommending TV show,and cold start of TV show cannot be solved.Therefore,a rating prediction algorithm is proposed for recommender system of live TV.This method adopts clustering technology to divide a day into some viewing timeslots for each account with different result so as to distinguish preferences of multiple users,proposes a strategy to cope with cold start of TV show,and constructes a rating function to predict user's preference.Thirdly,aiming at multi-viewer sharing account problem,the existing methods of recommending TV channel usually adopt timeslots to distinguish user's preferences,but it could not reflect the differences between each account,and the recommendation accuracy is low.Therefore,a time-aware recommender system of live TV is proposed.Basing on the method of recommendation TV channel,it improves the performance of algorithm by making use of time information.Each physical channel is divided into several virtual channels by clustering,and the users' preferences are mapped to different virtual channels after constructing user-virtual channel preference matrix,so this mehtod can handle cold start of TV show.The performance of this model is improved by introducing forgetting function,collaborative filtering technology and dynamic recommendation mehtod.Finally,a recommendation algorithm based on deep learning is proposed to utilize positive feedback and negative feedback respectively.A strategy to distinguish the positive feedback and negative feedback of live TV viewing behavior is proposed.Seven key characteristics are generated by dividing viewing timeslots,utilizing negative feedback,capturing continuous viewing preference and introducing the proportion of remaining time of candidate TV show.These features are not related with users' rating(feedback)of candidate at the recommendation moment,so this method can cope with the cold start of TV show.By taking recommendation of live TV as a classification task,neural network is adopted to integrate the above features to utilize users' feedback effectively and impove recommendation effect.