Research And Application Of Recommendation Strategies For IPTV Home Group Users

Compared with the traditional recommendation system business scenario,in the interactive Internet TV(IPTV)recommendation scenario,the users are mostly family group users with multiple members.The user behavior data is highly complex,and traditional recommendation algorithms cannot extract family members from it Interest preference information,it is difficult to accurately meet the recommendation needs of each member.In response to this problem,most of the existing research work starts with time information,uses clustering methods to establish the correspondence between time periods and family members,mines family members' interest preference information,and finally adopts a merge strategy to integrate the recommendation results of each member.Recommended list.This thesis aims to improve the accuracy of the recommendation algorithm in the IPTV home group user scenario,and proposes a new problem-solving idea.Each family member has their own preferred viewing period and interest preferences due to gender,age,work schedule,etc.,and will not easily change in a short period of time.Therefore,compared with single-member users,the time-based interest function of family group users tends to a periodic function,and the interest fluctuations within the cycle are larger due to the switching of viewing members.Based on the above phenomenon,this article considers family group users as single-member users whose interest preferences change on a day-to-day basis and the interest fluctuations within the period are large,capturing the fine-grained time effect of user interest changes as family members switch in one day.In the end,a recommendation list that changes with time for the family group user is provided to meet the recommendation needs of each family member at all times.The research process mainly completed the following tasks:1)Extract the hidden feature vectors of users in the home group at various periods in the cycle.The high-order singular value decomposition is performed on the scoring tensors based on the three dimensions of the user,time period,and item for the users in the home group to obtain the feature vector of the user in each time period.2)A time-sensitive hidden semantic recommendation model based on periodic function FamilyTimeSVD is proposed.Each bias and user factor in the implicit semantic model is established as a periodic function of time,and an attenuation factor is established based on the similarity of the periods to simulate the law of user interest changes over the period.3)A virtual family group user data set is built on the basis of the real single-member user data set,which solves the problems of lack of multi-member user scene data sets and lack of reference for experimental results.In terms of experiments,this thesis uses the Guizhou Broadcasting and Television real dataset and the Movie Family dataset to build a virtual family group dataset.The two datasets are compared with the TimeSVD ++ model.The experimental results show that the FamilyTimeSVD algorithm proposed in this thesis is more adaptable to multi-member user recommendation scenarios,the recommendation accuracy is higher,and it can better meet the precise recommendation needs of family members.