Research On Tag Recommendation And Group Fusion Algorithm Based On Implicit User Data

In this era of information overload,the traditional broadcast television platform recommendation algorithm can solve the problem of user program optimization to a large extent.Compared with the traditional TV recommendation algorithm,the existing TV program recommendation algorithm is aimed at the user's historical viewing record,but relying solely on the historical record does not fully satisfy the user's recommendation requirement.It is necessary to update the user preference attributes in real time based on the user's immediate selection.The existing TV recommendation algorithm is to default the viewing of the home user to the viewing behavior of the individual user.According to the TV usage habits in the daily household,the viewing record of the television is not only for the individual user,but also the family user,the current television.The study of the recommendation problem has hardly considered the recommendation of multi-user ratings.Therefore,this paper studies the recommendation problem of the hidden viewing behavior of broadcast TV users.The specific research contents are as follows.Firstly,aiming at the problem that user preferences can't be acquired explicitly in the process of TV program recommendation,this paper proposes a user preferences modeling method based on users' implicit viewing data.This method combines the viewing time of the user with the attribute of the program tag,builds the user's preference for the TV attribute tag,and establishes the user tag library vector for a long period of time.Considering the variability of user preferences,this paper designs a real-time recommendation feedback principle.According to this principle,when recommending in real-time,the feedback coefficient is obtained by using the difference between the ranking position of the program selected by users and the baseline,so as to update the tag library of users' preferences.Based on the above optimization and improvement,a label recommendation algorithm based on invisible user data is constructed.Secondly,aiming at the problem of home-based user recommendation with diverse interests,this paper proposes a multi-group fusion recommendation algorithm.This method calculates similarity by using user's viewing habits and viewing content,clusters multiple groups by K-means method according to similarity size,and then chooses appropriate fusion strategy to make the preferences of target users and K users nearest to target users.The preference for group integration.Finally,according to the two methods proposed in this paper,the real audience data design experiment of Qinhuangdao Radio Station is compared with the traditional userbased collaborative filtering algorithm.The label recommendation algorithm based on implicit viewing data is compared.The effect of the group fusion recommendation algorithm based on the viewing data on the hit rate;the comparison of the recommendation list diversity after the group fusion;the contribution of the three different fusion strategies to the algorithm.