Research On The Group Recommendation Method Based On Context-aware For Broadcast Television

A broadcast television user is a household composed of multiple members,whose watching interest is complex and corresponds to multiple watching groups.It is difficult to get accurate recommendation if TV recommender system is based on recommendation methods of web video.Generally,the watching interest of a TV user varies with viewing time,so the problem caused by composite user interest can be solved effectively by using context-aware recommendation.Considering that there are specific watching groups corresponding to specific time periods,so we can utilize time context to identify watching groups,and then to improve the accuracy of recommender system.The concept of group in existing group recommender system mainly means single household user,while few research work concentrated on the problem of group recommendation related to group which is composed of multiple household.For this problem,we need to solve two critical issues.The first issue is that we need to confirm TV users with composite similar watching interests,and the second issue is to obtain watching interests of group.In order to solve the problem of TV group recommendation,we need to obtain the features of programs and watching time to split watching interest which is composite,and then detect user group with similar watching interests to confirm recommended object.Finally,it is necessary to get watching interests of group to determine recommendation results.So this paper researches the problem of group recommendation aiming at broadcast TV users whose watching interests is composite,and complete the following work:1)Finishing the process of obtaining latent factor vectors of TV programs and watching time.First,calculating rating tensor according to history viewing records,and then grabbing latent factor vectors by using tensor decomposition.2)By splitting continuous time into several discrete time slots,a group recommendation method is achieved.Concretely,in order to detect TV group,we cluster programs and watching time slots according to their latent factor vectors,and then classify those records with similar time and program attributes.Finally,determine user group with similar watching interests.3)Implementing the recommended work for TV user group.Firstly,adopt a content-based method to calculate single user interest.After that,consider each user's enthusiasm in watching TV and adopt strategy of weighted multi-fusion preference to obtain the group's viewing preference.Finally,generate a recommendation list for the group.At last,experiments are performed on the broadcast TV dataset to verify the effectiveness of proposed TV group recommendation method.The experiments show that the proposed method reduces the average absolute error of group and any of the household belong to it in watching interest,and improves the similarity of watching interest of the members in the group.