Research On Personalized Recommendation Method For Online Media

With the rapid development of the Internet,the problem of information overload on the Internet has gradually become a focus of research.In order to help consumers to pick out information that is valuable to them from the chaotic information,and to help the producers of information to convert information into benefits more effectively,the recommendation system emerged.While the recommendation system brings convenience to people,there are still some problems that need to be resolved.This article focuses on the cold start problem,popularity bias problem,and data sparseness of the recommendation system.The main research results are as follows:For the cold start problem of the recommendation system,a cold start data configuration method of the recommendation system based on the multi-arm gamble machine model is proposed.First,the gambling machine model is initialized,according to the user set the specific parameters of the gambling machine;then,according to the user's information,the user's expectations are calculated and analyzed,the comprehensive calculation results for the user to select the recommended strategy;Finally,according to the user's feedback update The gambling machine model is ready for the next round of recommended strategy choices.Experimental verification shows that this method has certain feasibility and effectiveness.For the popularity bias problem of the recommendation system,a popular bias processing method for the recommendation system based on collaborative filtering and user opinion was proposed.First,the user's opinion is determined according to the user's information;after that,according to the user's opinion,the confidence function of the product popularity is constructed;finally,the confidence function is integrated with the similarity calculation in the collaborative filtering algorithm to complete the recommendation to the user.Experimental verification shows that the algorithm introduces user opinions into the collaborative filtering algorithm,which effectively mitigates the popular bias problem of the recommendation system.For the data sparsity problem of the recommendation system,a sparse data configuration method of the recommendation system based on convolutional neural network is proposed.First,according to the data and the two-dimensional characteristics of the convolutional neural network,the eigenvectors are constructed;after that the network structure of the convolutional neural network is established;then,the convolutional neural network is subjected to forward propagation training and reverse error adjustment;finally,produce a list of recommendations according to the results.Experimental verification shows that the algorithm is feasible and effective,and it has significant improvement in algorithm efficiency.A personalized recommendation system based on hybrid recommendation algorithm is designed and implemented.The recommendation algorithm studied in this paper is embedded in the recommendation system to achieve personalized recommendation for the user.The actual operation of the system shows that the recommended system designed in this paper works well.