Research And Application Of Personalized Recommendation In The Precision Marketing Of TV Programs

The convergence of the Broadcast TV Network and the Internet has enabled the sharing of TV program resources.A large number of TV programs have met the user's viewing needs,but it has also made it difficult for users to find programs of interest in interactive viewing options.How to realize the precision marketing of TV programs to meet the needs of users and increase the revenue has become a problem to be solved by TV program providers.The personalized recommendation system to deal with information overload can provide users with personalized services,but the personalized recommendation system in the field of TV programs still facing many problems.Aiming at the problem of no explicit rating and sparse data in the TV program recommendation system,this paper users Python language to analyze the historical behavior data of users and establish a personalized recommendation model.The main research contents are as follows:(1)Construct a linear weighted mapping function to form an implicit rating system.After analyzing the historical behavior data of users,it changes the traditional method of calculating ratings based on a single attribute,and proposes a weighted mapping function ratings system that combines multiple feature attributes in user behavior data.The characteristic attributes include:viewing time,number of views,payment amount,and number of payment,to obtain the user's the implicit rating matrix solves the problem of no user explicit rating in the TV program recommendation system.(2)Research on TV program recommendation system based on KNN collaborative filtering algorithm.Look for similar neighbors through the Cosine similarity coefficient formula,and make personalized recommendations based on the neighbors as the target user;and use the classification accuracy to compare the advantages and disadvantages of the recommendation system based on the User-CF algorithm and the Item-CF algorithm in the case of different neighbors number.The experimental results show that when the number of neighbors is 30,the indexes of the two algorithm models reach the best.And the Item-CF algorithm is more suitable for TV program recommendation system.(3)Research on TV program recommendation system based on matrix factorization algorithm.Establish a random gradient descent regression training model to minimize the loss function;consider the impact of user and program rating feature values and implicit information on the rating,add an offset term to the preliminary rating formula to optimize the model;and thenchange the number of hidden factors to find the minimum loss function.The experimental results show that when the number of hidden factors is 20,the loss function reaches a minimum value of about 0.27,and the Matrix Factorization algorithm is about 8% higher than the recommended model evaluation index F1-Score based on the KNN collaborative filtering algorithm.